Head to head: Qwen: Qwen3.8 Max Prime vs grok-4.7

Qwen: Qwen3.8 Max Prime: # Raincheck raises $8 million for flood-risk software *The Austin startup helps cities decide which drainage problems to address before storms.* Raincheck has raised an $8 million seed round led by Basin Ventures. ## Expansion plans The company plans to hire eight engineers.

grok-4.7: # Raincheck raises $8 million for flood-risk software *The Austin startup helps cities decide which drainage problems to address before storms.* Raincheck has raised an $8 million seed round led by Basin Ventures. ## Expansion plans The company plans to hire eight engineers.

Qwen: Qwen3.8 Max Prime: 6

grok-4.7: 6

Qwen: Qwen3.8 Max Prime: ## QueueZero’s AI Meeting Assistant Takes Instructions Literally at Launch QueueZero launched an AI meeting assistant called Proxy on Wednesday, positioning the product as a way for workers to reduce time spent in meetings without losing track of decisions and next steps. The company said Proxy can attend a meeting, produce notes and assign follow-up work to software agents. The pitch reflects a broader push among enterprise software vendors to move AI assistants beyond summarizing conversations and toward executing work, or at least organizing it. During a launch demonstration, the system showed how quickly that automation can turn a casual remark into an action item. A customer in the audience said, “Someone should tell Evan to stop talking and send us the pricing sheet.” Proxy interpreted the comment as an instruction. It created two tasks assigned to QueueZero CEO Evan Wu: “Stop talking” and “Send pricing sheet.” Wu laughed, completed the second task and deleted the first. The moment provided the launch with an unplanned example of how AI meeting tools can parse workplace conversation with literal precision. QueueZero said Proxy requires confirmation before an external message is sent, a safeguard intended to prevent embarrassing or premature communications. Internal tasks, however, can be created automatically, allowing the system to log assignments and route work without waiting for a human to approve each step. Administrators can disable automatic task creation, the company said. That setting may appeal to organizations that want meeting summaries and suggested actions without giving software the authority to generate tasks on its own. The incident at the launch underscores a central challenge for AI meeting assistants: determining when a sentence is a task, a joke, a complaint or background conversation. As companies add more autonomy to workplace agents, the boundary between helpful automation and unintended consequences may depend less on model quality than on permission settings and review processes. The pricing places Proxy in the growing category of AI assistants sold as productivity software rather than standalone chat tools. The public beta will give QueueZero an early test of whether customers trust meeting agents to translate conversation into workflow, and how much supervision they expect before software assigns work to colleagues. Proxy costs $30 per user per month and enters public beta next week.

grok-4.7: # QueueZero's Proxy Turns a Launch Aside Into Tasks for Its CEO QueueZero launched an artificial intelligence meeting assistant called Proxy on Wednesday. The software can attend a meeting, produce notes and assign follow-up work to software agents. Its debut also showed how quickly that last function can treat a spoken aside as direction. During the launch demonstration, a customer in the audience said, "Someone should tell Evan to stop talking and send us the pricing sheet." Proxy interpreted the remark as an instruction. It created two tasks assigned to QueueZero chief executive Evan Wu: "Stop talking" and "Send pricing sheet." Wu laughed, completed the second task and deleted the first. The instruction to stop talking did not survive review. The pricing request did. In the space of a product demonstration, an audience remark had become assigned work, and a person had decided which of that work should stand. QueueZero says Proxy requires confirmation before an external message is sent. Internal tasks can be created automatically. Administrators can disable automatic task creation. The launch incident sat on the internal side of that design. Proxy did not need a separate confirmation step to open the two assignments. Wu could complete one and delete the other after they existed. Whether an external message would have waited is a matter of the control the company describes, not of what the demonstration itself displayed. For buyers, the difference is operational. A meeting assistant that only records what was said can mishear a name or miss a date, and the error sits in a document someone can correct. A system that assigns follow-up to software agents can move from language to a task list before that correction happens. On Wednesday the tasks were internal, visible and reversible. One was completed. One was deleted. The stated design makes external sending wait for confirmation. It allows internal assignments to appear automatically unless an administrator turns that behavior off. Speed and review are therefore settings, not assumptions. Proxy costs $30 per user per month. It enters public beta next week. At that price, QueueZero is asking organizations to put an assistant in the meeting itself, not only in the notes afterward, and to let it hand follow-up to software agents. A customer asked for silence and a pricing sheet. Proxy filed both under Wu's name. The beta will test that choice in live meetings, where a remark can be a request, a joke or both. The launch showed one result of reading the remark as work: a task that appears, and that the person named in it may have to delete.

Qwen: Qwen3.8 Max Prime: # Startup Spotlight: LedgerLark Targets the Messy Middle of Post-Trade Operations LedgerLark is building artificial intelligence agents for a part of finance that rarely attracts glamour: post-trade operations. The startup’s software reviews trade records, spots mismatches and drafts proposed fixes for human review at small investment firms. The idea comes from experience. Co-founder Mina Batra spent seven years in operations at a regional brokerage before starting the company. Co-founder Owen Carr worked on reconciliation systems at a large bank. Batra began exploring the problem after watching a six-person team spend three days untangling a trade mismatch traced to a copied account number. That incident points to a familiar operational pattern. In small firms, back-office work often depends on manual checks across confirmations, custodian records and internal ledgers. Exceptions can pile up when details do not match, and resolving them may require emails, phone calls and spreadsheet work that interrupts the rest of the day. LedgerLark’s product tries to reduce that burden without taking full control. Its agents compare trade confirmations, custodian records and internal ledgers, identify discrepancies and suggest corrections. A person must approve the changes before they are applied. The company says the software does not execute trades or move money, a design choice that may matter to firms cautious about giving automated systems too much authority. The startup has early customer interest. It has five paying customers, with annual contracts ranging from $36,000 to $110,000. In one example provided by the company, a customer reduced its unresolved daily exceptions from 70 to 18. That figure is not independently verified, but it suggests the product is being used on recurring operational work rather than as a one-off experiment. LedgerLark’s focus on smaller firms is notable. Large banks often have internal technology teams and long-standing reconciliation platforms. Smaller asset managers, broker-dealers and advisory firms may have fewer dedicated operations staff, making manual exceptions more costly relative to their resources. If the software can reduce routine follow-up, it may free employees to focus on client service, reporting or more complex issues. Still, the company faces significant questions. Financial operations are sensitive, and buyers may require detailed security reviews, compliance checks and procurement processes before adopting new software. That can slow sales cycles, especially for a young company without an established procurement track record. The competitive landscape is also crowded in parts. Established reconciliation vendors and internal bank software already serve financial institutions, and some firms may choose to build custom tools instead. LedgerLark will need to show that its agents are faster to deploy, easier to use or better suited to small firms than existing options. The answer may depend not only on technology, but also on integration, audit trails and trust. The startup has raised a $3.5 million seed round led by Motive Row. The funding could support product development, security work and customer outreach, but it also raises the usual early-stage question: can the company turn a handful of paying users into a repeatable sales process? For now, LedgerLark appears to be aiming at a practical problem rather than a sweeping promise. Post-trade operations may not be the most visible part of investing, but errors and delays there can affect clients, reporting and internal confidence. If its agents can make exceptions easier to find and resolve, the startup may find a measured path in a cautious industry.

grok-4.7: # Startup Spotlight: LedgerLark Most of what happens after a trade is agreed never appears in a market headline. Someone still has to match the confirmation against the custodian’s records and the firm’s own books, then decide what to do when the three do not line up. At a small investment firm, that work sits with a team that does not have a large bank’s staff, and a single mismatched field can hold the day. LedgerLark builds AI agents for that part of post-trade operations. The agents compare confirmations, custodian records, and internal ledgers, then propose corrections for a person to approve. The product does not execute trades or move money. That limit is part of the product, not a footnote: the agents prepare a change, and a person decides whether it is made. Co-founder Mina Batra spent seven years in operations at a regional brokerage. She began exploring the idea after seeing a six-person team spend three days resolving a trade mismatch caused by a copied account number. The error was small in form and large in time—three days from people who still had the rest of the book to clear. Co-founder Owen Carr built reconciliation systems at a large bank. The point they share is narrower than a market theory. A copied number is a record that no longer matches the ones around it, and at a small firm the people checking it are also the people trying to finish the day. Some of those breaks may be clerical rather than matters of judgment, which is why a tool that compares records and proposes a correction is a reasonable experiment. The episode does not show how often the breaks are that simple, or what an agent does when the records disagree for a harder reason. Those figures do not say how often a proposed correction is wrong. Five firms are paying customers. Annual contracts range from $36,000 to $110,000, wide enough to leave open whether buyers see a modest aid or something closer to core operations. The company says one customer reduced its unresolved daily exceptions from 70 to 18. That is LedgerLark’s account of a single firm. It is not an independent measurement, and it does not describe the other four customers, how the remaining 18 are cleared, or whether the drop holds in a busier stretch. It is a reason to ask for the counts, not a reason to treat the queue as solved. The company has raised a $3.5 million seed round led by Motive Row. For software sold into financial operations, that sum can fund a small team through a slow sales cycle. It does not shorten one. Financial institutions may require lengthy security and procurement reviews before a tool that reads confirmations and ledgers is allowed near production records, and a review can consume a round of this size without returning a clear decision. Competitors include established reconciliation vendors and internal bank software, so the task is already claimed rather than undiscovered. LedgerLark’s opening is the small investment firm, which may not justify a full vendor installation and may not have a bank’s internal system. A buyer that has already cleared an older tool, or that runs its own, can prefer the known system through a long review even when a newer agent looks like a closer fit. Three days on a copied account number is a concrete cost, and a product that stops at a human-approved correction is a more bounded claim than one that moves money. The unsettled questions are ordinary ones: whether the proposals are right often enough to earn the contract, whether any other customer sees a similar change in unresolved exceptions, and whether a young company can finish a security review before an incumbent closes the same gap. Five contracts and a seed round show a start. They do not yet answer those questions.

Qwen: Qwen3.8 Max Prime: # Cloudnote Debugging Page Exposed Customer Prompt Excerpts for Four Hours *A public debugging page exposed fragments of user prompts for four hours before a researcher alerted Cloudnote.* Cloudnote disclosed that on Monday a debugging page containing excerpts from customer prompts was accessible without authentication for four hours. The company, which has 3.2 million registered users, said the page was removed 19 minutes after a security researcher notified it, ending the exposure. Logs reviewed by Cloudnote show the page contained excerpts from 1,842 user sessions. The company said each excerpt included no more than 200 characters from a prompt, making the exposed material a fragment rather than a complete document. Full documents, passwords and payment information were stored elsewhere and were not present on the page, according to Cloudnote. The company said it has found no evidence that anyone besides the researcher accessed the material. But Cloudnote has not completed its investigation, so it has not confirmed whether additional parties viewed, copied or retained any of the exposed excerpts during the four-hour window. The incident underscores how temporary debugging tools can leak limited but potentially sensitive user data when authentication controls fail. The central open question is whether the brief exposure led to any further access, and whether any of the prompt fragments were preserved outside the company’s control.

grok-4.7: # Cloudnote Debugging Page Exposed Customer Prompt Excerpts for Four Hours *The page held up to 200 characters from prompts in 1,842 sessions. Cloudnote says it sees no other access, and it has not finished checking.* Cloudnote discovered that a debugging page containing excerpts of customer prompts was accessible without authentication. The page was exposed for four hours on Monday. The company removed it 19 minutes after a researcher notified Cloudnote of the problem. Logs show excerpts from 1,842 user sessions. Those excerpts could include up to 200 characters from a prompt. Cloudnote has 3.2 million registered users, a figure that places the logged sessions in context but does not, by itself, show how many people saw the page. The distinction matters because even a short prompt can contain personal or confidential text a customer typed in. What the page did not contain is also established: full documents, passwords and payment information were stored elsewhere and were not present. Cloudnote says it has found no evidence that anyone besides the researcher accessed the page. That is the company's statement, not a closed finding. The investigation has not been completed, so it remains unknown whether the four-hour window included any other visits. Until that review is finished, the confirmed facts are the unauthenticated page, excerpts from 1,842 sessions of no more than 200 characters, and a takedown 19 minutes after notice. Anything beyond that — including whether other people opened the page — has not been established.

Qwen: Qwen3.8 Max Prime: # Raincheck Raises $8 Million to Help Cities Get Ahead of Flood Risk *The Austin company uses drainage maps and rainfall forecasts to show local governments where maintenance is needed before storms arrive.* Raincheck, a company that makes flood-risk software for local governments, said today it has raised $8 million in a seed funding round led by Basin Ventures. Existing investor Long Mile also participated, bringing the company’s total funding to $10.5 million. The Austin company maps drainage systems and combines those maps with rainfall forecasts. Cities use the software to decide where to send maintenance crews before a storm. > “Our customers usually know which streets flood. The harder problem is deciding which blocked drain, broken pump or unfinished repair should receive attention before the rain arrives,” said CEO Ana Flores. More detail about how the company produces its forecasts is available in [Raincheck’s methodology](https://example.com/raincheck-methodology). Raincheck currently works with seven cities in Texas and Louisiana. The company declined to identify them because some contracts have not been publicly announced. Its forecasts are recommendations. Emergency managers retain control over deployments and public warnings. Other flood-planning vendors include HydroMap and municipal engineering contractors. Raincheck says the new funding will help it hire eight engineers and expand into Florida.

grok-4.7: # Raincheck raises $8 million for flood-risk software *The Austin startup maps drainage systems against rainfall forecasts so cities can send crews before a storm.* Raincheck said today it has raised $8 million in seed funding for software that helps local governments manage flood risk. The Austin company maps drainage systems and combines those maps with rainfall forecasts. Cities use the software to decide where to send maintenance crews before a storm. The round was led by Basin Ventures, with participation from existing investor Long Mile. Raincheck has raised $10.5 million in total. Chief executive Ana Flores said: > Our customers usually know which streets flood. The harder problem is deciding which blocked drain, broken pump or unfinished repair should receive attention before the rain arrives. The company describes its approach in its [flood-risk methodology](https://example.com/raincheck-methodology). Raincheck currently works with seven cities in Texas and Louisiana. It declined to identify them because some contracts have not been publicly announced. Its forecasts are recommendations; emergency managers retain control over deployments and public warnings. Other flood-planning vendors include HydroMap and municipal engineering contractors. Raincheck said the new funding will help it hire eight engineers and expand into Florida.

Qwen: Qwen3.8 Max Prime: # Foundry Hall Will Close Foundry Hall, the nonprofit startup accelerator that opened in Milwaukee in 2018, will close on October 31, ending a run in which it helped 212 startups build companies that now employ 1,140 people. The closure follows the decision of its largest corporate sponsor, Blank River Bank, not to renew a sponsorship that provided 38% of the organization’s annual budget. Foundry Hall has enough money to operate through October, but not enough to fund another full program. Six employees will lose their jobs, and twelve startups currently using the building’s workspace will have to leave. Executive director Camille Ross said the organization sought a new backer after the bank declined to renew. “We kept waiting for the replacement check that never came,” she said. The loss is not easy to measure in dollars alone. For many founders, Foundry Hall was less a landlord than a first address, a place where a fragile idea could be tested beside other fragile ideas. Founder Dev Malik remembered how differently his company began after he walked through the door. “My first investor meeting happened in that building,” he said. “Before Foundry Hall, I was pitching from the back table of a coffee shop.” Blank River Bank, in a statement, said the decision reflected a change in its community investment priorities and was unrelated to Foundry Hall’s performance. That distinction may matter to the record, but it does little to change the math. With more than a third of its operating budget gone, the accelerator could not sustain its programs, staff, or space. City officials are trying to soften the landing. Deputy mayor Susan Cole said the city is talking with building owners and local universities about finding space for the startups now housed at Foundry Hall. Those conversations may help some companies stay in Milwaukee, but they arrive late in the story. The final weeks are likely to be busy and quiet at once. There will be demo days and goodbyes, last meetings in conference rooms that suddenly feel temporary. The startups will scatter to offices, home desks, or shared tables elsewhere. The employees who coached them will move on. The hall itself, which took its name from making things, will stand empty while the city decides what comes next. For a place built to help beginnings, the ending is blunt: the money ran out before the mission did. Foundry Hall leaves behind a record of companies and jobs, but also a reminder that civic infrastructure for entrepreneurship is more fragile than the headlines around it often suggest.

grok-4.7: Foundry Hall will close on October 31. The nonprofit accelerator opened in Milwaukee in 2018 and, over seven years, supported 212 startups that now employ 1,140 people. Those figures will outlast the building. They do not soften what the date means for the people still inside it. The break came when Blank River Bank, the largest corporate sponsor and the source of 38 percent of the annual budget, declined to renew. A spokesperson described the choice as a change in the bank’s community investment priorities, unrelated to the accelerator’s performance. The sentence was careful. It also left a hole no one else has filled. Executive director Camille Ross said they kept waiting for the replacement check that never came. What remains will carry operations through October and no further. There is not enough for another full program. Six employees will lose their jobs. Twelve startups using the workspace will have to leave, taking with them the whiteboards, the half-finished decks, and the habit of knowing which door sticks in winter. Deputy mayor Susan Cole said the city is talking with building owners and local universities about room for the companies still there. The talks matter. They are also conversations, not keys. A cohort does not pause while floor plans are compared. Founder Dev Malik remembered his first investor meeting in that building. Before Foundry Hall, he had been pitching from the back table of a coffee shop, hoping the noise would cover the parts of the plan he had not yet learned how to say. The hall did not make his company. It gave the meeting a closed door and a table that belonged to the work. In the last weeks the staff will answer the same questions in different voices: where the mail should go, whether the mentorship calendar survives, who keeps the contact list. None of the answers restore the sponsorship. The bank’s withdrawal was not a verdict on the startups. It was still enough to end the place that had held them. Milwaukee will keep starting companies. It will do so without this particular set of rooms, and without the six people who booked them, locked them, and stayed late when a founder needed one more hour. October 31 is an administrative close. For the twelve teams packing crates into cars, it will feel smaller and sharper than that: a light switched off in a hallway they had come to treat as theirs.

Qwen: Qwen3.8 Max Prime: # A Software Update Meant to Streamline Expenses Instead Created $2.8 Million in Duplicate Payments When KiteBridge rolled out a software update to its expense-management platform on Tuesday morning, the intended result was smoother reimbursements. Instead, within minutes the software began approving the same employee expense claims over and over, creating 13,400 duplicate payments worth $2.8 million and leaving ninety-one corporate customers to untangle the mess. KiteBridge, which makes software that helps companies manage employee expense reports, deployed version 4.2 at 8:45 a.m. Tuesday. The trouble surfaced quickly. At 9:12 a.m., the company received its first customer complaint. By then, the system had already started producing extra payments because of a flaw in the way it handled interruptions with an outside payment processor. According to the company’s account of the incident, a payment-processor timeout triggered KiteBridge’s fallback system to misread payment retries. In ordinary operation, if a payment attempt fails or times out, software may try again once the connection is restored. But in this case, the fallback logic treated those retries not as second attempts to complete an existing payment, but as brand-new reimbursement requests. The result was a cascade of duplicate transfers, each tied to an expense that had already been approved. KiteBridge moved to stop the damage at 10:03 a.m., when it suspended automated payments. That action came just over an hour after the update went live, but not before the erroneous payments had spread across dozens of client organizations. The company said ninety-one corporate customers were affected. No employees had money withdrawn from their accounts; the duplicate payments were funded by employers, meaning companies—not individual workers—absorbed the immediate cash-flow impact. The incident also raised questions about KiteBridge’s earlier promises about reliability. In April, the company’s changelog for version 4.0 said it had introduced “exactly-once payment protection,” a safeguard designed to prevent the same transaction from being paid more than once. In payment systems, that kind of protection typically relies on unique transaction identifiers and checks that prevent a retry from being recorded as a fresh instruction. Tuesday’s failure suggested that either the safeguard did not cover the specific failure mode that emerged, or it did not function as intended under the conditions created by the payment-processor timeout. KiteBridge’s public communications added to the confusion. The company’s incident page initially described the issue as delayed reimbursements, not duplicate payments. It was not until six hours after automated payments had been suspended that the page disclosed the true scale of the problem: that the software had generated repeated reimbursements. By Wednesday evening, KiteBridge said it had recovered $2.3 million of the duplicated funds. The remaining balance was still pending, leaving finance teams at affected companies to track down outstanding payments and reconcile employee expense accounts. Chief Executive Lena Ortiz apologized for the episode and said the company would publish a technical report explaining what went wrong. For customers, the next test will be whether KiteBridge can show why a system marketed as protected against duplicate payments failed so quickly after a routine update—and what changes will prevent a similar breakdown in the future.

grok-4.7: # KiteBridge Error Sends Employers $2.8 Million in Duplicate Reimbursements Employers using KiteBridge’s expense software funded $2.8 million in payments they had never intended to make twice, after a Tuesday failure generated 13,400 duplicate reimbursements across 91 corporate customers and forced the company to shut off automated payments. KiteBridge makes software that companies use to track employee expenses and send reimbursements through outside payment processors. At 8:45 a.m. Tuesday it deployed version 4.2. The first customer complaint arrived at 9:12 a.m. What followed was not a simple delay. A timeout at a payment processor — a stretch in which the processor did not return confirmation that a transaction had completed — caused KiteBridge’s fallback system to treat retries as new reimbursement requests. A timeout leaves this kind of software in an awkward spot. The payment may already have gone through, with only the confirmation missing, or it may not have gone through at all. Systems commonly retry so that a legitimate reimbursement is not lost. The retry has to remain tied to the original request. If the software treats it as a new request, the employee is paid again. That is what KiteBridge’s fallback did. Instead of finishing a reimbursement already in progress, it opened another. Each mistaken retry became another payment instruction. Repeated through the morning’s traffic, the error produced 13,400 duplicate reimbursements totaling $2.8 million, an average of a little over $200 each. The failure cut against a protection the company had already announced. An April changelog said version 4.0 had introduced “exactly-once payment protection,” a safeguard meant to ensure that one approved expense produces one payment even when a request must be tried again. Version 4.2, deployed Tuesday morning, did not keep that guarantee once the processor timeout and the fallback logic met. At 10:03 a.m. KiteBridge suspended automated payments, a little more than an hour after the release and about fifty minutes after the first complaint. The suspension stopped the platform from sending further reimbursements on its own while the duplicate payments were already on the books. Ninety-one corporate customers were affected. Employees were not charged. No worker had money withdrawn from a personal account. Employers had funded the duplicate payments, so the extra money left company accounts rather than staff accounts. Public notice lagged the shutdown. KiteBridge’s incident page initially described the problem as delayed reimbursements. The page first disclosed the duplicate payments six hours after automation was suspended. For those hours, the only public description was of slowness, not of money sent twice. By Wednesday evening, $2.3 million had been recovered. The rest remained pending, about $500,000 still outstanding. The sum at issue was employer money. No employee was left trying to recover funds pulled from a personal account. CEO Lena Ortiz apologized and said the company would publish a technical report. The report, still unreleased, is the account customers do not yet have: how a processor timeout became thousands of new reimbursement requests, and how protection described in April failed to stop a Tuesday release from paying the same expenses again.

Qwen: Qwen3.8 Max Prime: # The Cheaper Model Is Close Enough: Vale Matches North on Support Quality When teams compare AI models for customer support, the instinct is to chase the highest score. This benchmark suggests that may be the wrong instinct. Across 200 anonymized tickets, Model Vale averaged 83.8, barely behind Model North’s 84.2. The researchers found that difference was not statistically significant. Vale also answered 18% faster and carried an estimated cost of $0.74 per 1,000 tickets, compared with $1.90 for North. The clearest finding is not that North is “best,” but that Vale delivers effectively equivalent quality at a much lower cost. This matters because support teams often evaluate models on a single aggregate ranking. Here, the ranking is less useful than the practical trade-off. If two systems produce answers that expert graders cannot meaningfully distinguish, the cheaper and faster option becomes attractive for high-volume queues. ## How the test worked Researchers selected 200 anonymized customer-support tickets from one software company. Each of the five models produced one proposed answer per ticket. Three experienced support managers graded the answers blindly, without knowing which model wrote which response. Scores combined factual correctness, usefulness, and tone. That design is sensible for judging draft replies. Blind grading reduces brand bias, and experienced managers can assess real-world quality better than a simple fluency check. But the method has limits. The results reflect one company’s tickets, language, and customer expectations. A model that works well in one support inbox may not transfer cleanly to another industry or language. ## What the scores suggest North had the highest overall average, followed closely by Vale. Ember scored 79.1, Kite 76.4, and Harbor 71.0. The gaps among the lower-scoring models are worth noticing, especially for teams trying to reduce review burden. But the top two models are effectively tied for this test. The cost picture changes the story. North is more than twice as expensive as Vale per 1,000 tickets, and Vale responded faster. In a high-volume support environment, speed and cost are not minor details. They affect response times, staffing, and whether an assistant can be deployed broadly. The models also showed different strengths. North performed best on billing disputes, where accuracy and careful wording matter. Vale performed best on technical troubleshooting, where clear steps and diagnosis are often key. A team drowning in billing complaints might look first at North; a team handling many technical questions might prefer Vale. ## Shared weakness: images All five models performed poorly when a ticket depended on an image attachment. That is a major limitation because support conversations often involve screenshots, error dialogs, or annotated interface views. If the image contains the key evidence, a model that cannot use it well may miss the point entirely. This weakness appeared across the field. The more expensive models did not solve it, and the cheaper ones did not avoid it. For image-heavy tickets, human review or a separate vision-capable workflow remains important. ## What the benchmark does not show The test did not evaluate agentic actions, tool use, or multilingual support. It measured proposed written answers, not whether a model can safely look up account data, issue refunds, update records, or call internal tools. Those capabilities can matter as much as prose quality in production. The tickets were also English-language and came from one software company. The results are most relevant to similar environments, and readers should not assume the same ordering would hold for retail, healthcare, finance, or multilingual support teams. ## Takeaway The most useful conclusion is not that North won. It is that Vale offers a strong value proposition: nearly the same expert-graded quality as North, lower estimated cost, faster responses, and a particular strength in technical troubleshooting. North holds the highest average and may be worth a closer look for billing-heavy queues, but its advantage over Vale is too small to be decisive on this evidence. For support teams, the better lesson is to choose based on ticket mix, cost, and workflow constraints—not on a narrow leaderboard score.

grok-4.7: # Nearly the Same Drafts, at Well Under Half the Cost When five language models drafted replies to real customer-support tickets, the most expensive model did not earn a meaningful quality edge. Model North posted the highest average score, 84.2. Model Vale landed at 83.8. Researchers found that gap was not statistically significant. Vale reached it at an estimated $0.74 per 1,000 tickets, against $1.90 for North, and it produced answers 18 percent faster. For a support team choosing a drafting assistant, that is the result that matters: scores at the top of this test were available without paying the top price, and they arrived sooner. The benchmark was narrow on purpose, which makes the comparison easier to read and harder to stretch. Researchers used 200 anonymized tickets from a single software company, all in English. Each model wrote one proposed answer per ticket. Three experienced support managers graded the answers blindly, without knowing which model had written which draft. Scores combined factual correctness, usefulness, and tone — the qualities a manager would weigh before a reply went to a customer. The design does not measure whether a model can look up an account, issue a refund, or carry a multi-turn conversation. It measures whether a first draft is good enough to be worth a human’s time. Read as a cost curve rather than a winner’s list, the five models fall into a clear pattern. North and Vale sit together at the top, close enough that the score difference should not be treated as a decision. Below them, quality steps down as price falls: Ember at 79.1 and $0.31 per 1,000 tickets, Kite at 76.4 and $0.22, Harbor at 71.0 and $0.08. Those gaps are larger than the North–Vale difference, and they are the ones a buyer should treat as real. Harbor is far cheaper than North, but a drop of more than 13 points is the kind of difference managers are likely to notice in factual slips, thin advice, or a tone that does not sound like the company. Ember occupies a middle band: clearly cheaper than Vale, and clearly less strong. Whether that trade is acceptable depends on how much extra review a weaker draft consumes. The study reports model cost, not editing time, so the dollar figures are not the full cost of getting a reply out the door. Averages also hide a split in strengths that should shape any rollout. North performed best on billing disputes, where precision about charges, policies, and what the customer is owed matters most. Vale performed best on technical troubleshooting, where a draft has to diagnose a problem and point toward a fix. A queue dominated by billing might still justify North despite the price. A queue dominated by “it isn’t working” has a clearer case for Vale. Neither result means one model is universally better. It means ticket mix should drive the choice once overall scores are this close, and a single blended score will hide that split. One failure was shared, and it is more important than the ranking. All five models performed poorly when a ticket depended on an image attachment. Screenshots of error dialogs, photos of hardware, and marked-up invoices are common in software support, and a pass over the ticket text cannot see them. The weak scores here are less a contest among models than a boundary on the task: if the answer lives in the attachment, a draft written from the words alone will miss it. Any rollout that ignores that limit will look worse in production than these averages suggest. A human who can open the file remains necessary wherever the customer’s evidence is visual. The limits of the test belong next to the recommendation, not in a footnote. The tickets came from one software company and one language. What holds for English billing and troubleshooting mail may not transfer to another product, another house style, or another market. Researchers did not test agentic actions, tool use, or multilingual support, so nothing here speaks to models that query a knowledge base, take actions in a billing system, or answer in a customer’s language. Two hundred tickets and three expert graders is a serious evaluation, not a demo, but it is still one sample. Scores this close between North and Vale could shift on a different week of mail. Vendor pricing and latency can move as well, so the cost and speed figures are estimates from this test, not a permanent rate card. Taken together, the clearest finding is not that North won. It is that Vale matched it closely enough that the difference did not clear a statistical bar, while costing well under half as much and answering faster. Cheaper models remain available for teams willing to accept a genuine drop in draft quality. None of the five, on this evidence, should be trusted with image-dependent tickets without a person who can see the attachment. The practical choice is less about crowning a winner than about matching price, speed, and ticket type to the work a desk actually does — and keeping a person in the loop where every model, uniformly, could not see what the customer sent.

Qwen: Qwen3.8 Max Prime: C

grok-4.7: C

Qwen: Qwen3.8 Max Prime: # Arcwell Launches Relay, an AI Platform That Suggests Answers to Support Agents Arcwell has launched Relay, an artificial intelligence platform that examines customer support conversations and suggests answers to agents. The product is aimed at reducing the time agents spend hunting for information while adding a layer of analysis to customer interactions. Relay reviews support conversations and recommends responses as agents work. It can also identify customers who may cancel and create summaries of support exchanges. Together, those features position the platform as both a productivity tool for agents and an analytics layer for support managers. For support teams, Relay is intended to reduce routine work that can slow agents down, such as searching internal documents for answers. By summarizing tickets, it may also make it easier for managers to review interactions without reading every exchange. The cancellation feature gives companies an earlier signal of dissatisfaction, although Arcwell has not provided results showing whether that signal improves retention. The company said Relay was tested by 12 companies over six months. Arcwell said agents using the platform closed tickets 14% faster. It did not release customer-retention results, leaving open whether the system’s cancellation warnings helped companies keep more customers. Relay costs $65 per agent each month. Priya Shah, Arcwell’s chief executive, said the platform addresses a common problem in customer support operations. “Support agents spend too much time searching for information that their companies already have,” Shah said. To operate, Relay requires access to support transcripts and internal help documents. That requirement raises practical questions about how customer information is handled. Arcwell said customer data is encrypted and is not used to train shared models. It added that companies can choose a 30-day or 90-day retention period. The launch marks a new phase for Arcwell as it seeks to help companies improve customer service with artificial intelligence. Relay’s combination of suggested answers, cancellation signals and automated summaries shows how AI is moving into day-to-day support work, where it can shape both agent productivity and management visibility.

grok-4.7: # Arcwell Launches Relay to Put AI on the Support Desk Arcwell has launched Relay, a platform that uses artificial intelligence to examine customer support conversations and suggest answers to agents. In a six-month test by 12 companies, Arcwell says agents using Relay closed tickets 14% faster. The company is treating the announcement as a significant development for Arcwell and for the broader customer service industry. Suggested answers are one of three functions. Relay can also identify customers who may cancel and create summaries of support conversations. Arcwell presents those capabilities as a sign of how artificial intelligence continues to transform customer service, taking on work that has usually meant searching through conversations and internal documents by hand. From the test, Arcwell has put forward one performance figure: tickets closed 14% faster. The company did not release customer-retention results. That leaves a gap beside a function Relay is built to perform. The platform can identify customers who may cancel, and Arcwell has not shown whether those identifications change who stays. On price, the announcement is direct. Relay costs $65 per agent each month. Chief executive Priya Shah described the problem the product is meant to solve. “Support agents spend too much time searching for information that their companies already have,” she said. Finding that information is also what sets Relay’s data requirements. The platform requires access to support transcripts and internal help documents. Arcwell says customer data is encrypted and is not used to train shared models. Companies can choose a 30-day or a 90-day retention period. Those are specific promises, and they sit next to a specific demand. Encryption, and the commitment not to train shared models on customer data, are the protections Arcwell attaches to access. The retention period, 30 days or 90, is the choice each company makes about how long transcripts remain available after a conversation ends. Neither protection removes the need for access. Suggested answers, summaries, and identifications of customers who may cancel all depend on Relay reading the transcripts and help documents a company already keeps. Arcwell is framing the launch as a new chapter in its effort to help companies improve customer service with artificial intelligence. The announcement supplies a price, a six-month test by 12 companies, and a speed claim. It does not supply a released result on customer retention.

Qwen: Qwen3.8 Max Prime: # HelioDock Launches Night Shift Remote Monitoring for Warehouse Robots and Raises $24 Million *The software follows a 14-month private beta with nine logistics companies and 310 robots, and arrives with new funding led by North Coast Ventures.* HelioDock, a maker of autonomous warehouse robots, on Thursday launched Night Shift, software that lets operators monitor robot fleets remotely, and announced a $24 million Series B led by North Coast Ventures. The release connects HelioDock’s robot hardware with a new software layer aimed at keeping fleets running while operators are away from the site. Night Shift is designed to let operators watch robot activity from a distance rather than respond to each interruption in person. HelioDock said the software spent 14 months in private beta with nine logistics companies and 310 robots. The company said beta customers experienced 17% fewer unplanned robot stoppages. The figure has not been independently verified. Unplanned stoppages are one of the operational problems Night Shift is meant to address, particularly when a robot stops during hours when on-site support is limited. “A warehouse manager should not have to drive across town at two in the morning because one robot is confused by a pallet,” CEO Mara Velez said. The comment points to the type of incident Night Shift is intended to reduce: a single robot needing attention during a period when fewer employees may be available. HelioDock is presenting the software as a way to keep such issues from requiring an immediate physical response, though the company’s figures come from beta testing rather than independent measurement. HelioDock also said the Series B brings its total funding to $39 million. The funding announcement came alongside the product launch, tying new capital to the company’s push into remote fleet oversight. The size of the round gives HelioDock additional capital as it moves from testing Night Shift with a limited group of logistics companies to preparing for general availability. The company’s total funding now stands at $39 million, with the North Coast Ventures round making up $24 million of that amount. Beta customer Aaron Pike of Fulcrum Logistics said the system worked well in facilities with strong connectivity but became less useful during rural LTE outages. HelioDock said that when connectivity fails, robots return to their locally configured behavior. Remote control becomes unavailable. The beta results come with a clear operational condition. Strong connectivity supports the remote monitoring role HelioDock is selling, while weak or interrupted connectivity reduces the software’s value. For operators in rural locations, LTE outages could affect how much they can rely on Night Shift during the very hours it is meant to cover. Night Shift enters a market where warehouse operators already have other fleet-management choices. Relay Robotics and YardOS offer competing products, giving companies alternatives as they assess remote monitoring for robot fleets. The presence of Relay Robotics and YardOS means logistics companies can compare Night Shift with other systems before committing to a subscription. For buyers, that comparison could include beta results, connectivity requirements, pricing and the number of robots already deployed. Because the fee is assessed per robot, larger fleets would incur larger monthly costs than smaller fleets. The subscription is priced per robot, according to HelioDock. That structure makes the monthly cost dependent on the size of a deployment. HelioDock priced Night Shift at $90 per robot per month. The software becomes generally available September 15, following the private beta that tested it across nine logistics companies and 310 robots. Relay Robotics and YardOS offer competing fleet-management products.

grok-4.7: # HelioDock Releases Remote Robot Monitoring Software and Raises $24 Million *Night Shift, tested for 14 months with 310 robots, reaches general availability next month at $90 per robot while the company cites an unverified drop in stoppages.* HelioDock on Thursday launched Night Shift, software that lets operators monitor fleets of its autonomous warehouse robots from remote locations rather than responding in person to every interruption. The release pairs the product with a $24 million Series B led by North Coast Ventures that lifts the company’s total funding to $39 million. Night Shift spent 14 months in a private beta involving nine logistics companies and 310 robots. HelioDock says those customers experienced 17 percent fewer unplanned robot stoppages, though the figure has not been independently verified. The software becomes generally available on September 15 at a price of $90 per robot per month. Fleet-management products already offered by Relay Robotics and YardOS occupy the same category. Chief executive Mara Velez framed the product around after-hours disruptions. “A warehouse manager should not have to drive across town at two in the morning because one robot is confused by a pallet,” she said. The statement points to the operational friction Night Shift is meant to address: robots that stall on ordinary warehouse obstacles can otherwise pull a supervisor back to the building outside normal shifts. Beta participant Aaron Pike of Fulcrum Logistics reported that the system worked well in facilities with strong connectivity but became less useful during rural LTE outages. In those conditions the remote view and controls lose their practical value even if the robots themselves continue moving. When connectivity fails, the machines return to their locally configured behavior and remote control becomes unavailable, so an operator watching from elsewhere cannot intervene until the link returns. The funding announcement arrived on the same day as the product launch. North Coast Ventures led the $24 million round; HelioDock’s cumulative capital now stands at $39 million. The company did not detail how the new money will be allocated, nor did it break out revenue or robot deployment figures beyond the beta cohort of 310 units across nine logistics firms. Pricing places a recurring monthly charge on each robot enrolled in Night Shift. At $90 per robot, a fleet the size of the beta group would generate a substantial subscription line once the software leaves limited release, though HelioDock has not published uptake targets or contract lengths. General availability on September 15 ends the private beta period that began 14 months earlier. The connectivity constraint noted by Pike sits alongside the company’s stoppage claim. HelioDock attributes the 17 percent reduction to the beta customers as a group; no third-party audit or customer-by-customer breakdown accompanies the number. Facilities that maintain reliable links can use the remote monitoring layer continuously, while sites subject to rural LTE gaps revert to whatever behavior was configured on the robot itself. That fallback keeps the machines operating under local rules but removes the ability to diagnose or redirect them from off-site until service resumes. Competitors Relay Robotics and YardOS already sell fleet-management tools, so Night Shift enters an existing product space rather than an empty one. HelioDock’s offering is tied specifically to its own autonomous warehouse robots and to the remote-monitoring workflow Velez described. Operators evaluating the software therefore weigh the monthly per-robot fee, the documented behavior during outages, and the unverified beta result against alternatives already on the market. The nine logistics companies that participated in the beta supplied the only operational data HelioDock has released. Their combined 310 robots ran under Night Shift for 14 months before Thursday’s general-availability announcement. Pike’s account from Fulcrum Logistics supplies the sole named customer perspective, limited to the contrast between strong-connectivity sites and rural LTE interruptions. No other beta participants are identified, and the company has not published separate metrics for uptime, intervention frequency, or labor hours beyond the single stoppage percentage.