AI — Page 3
Models, agents, infra, applied AI.
- Head to head: Bernini-R Edit Image vs Happy Horse 1.1 Image to Video
This matchup turns on execution, not vibes. Bernini-R Edit Image flashes style, but Happy Horse 1.1 Image to Video is the model that more consistently obeys the brief, preserves scene logic, and delivers cleaner motion storytelling across both tests.
- Head to head: Bagel vs Krea 2 Large
This matchup splits cleanly between Bagel’s eye for composition and Krea 2 Large’s stricter obedience to the brief. One model makes prettier images when it gets room to improvise; the other wins by actually delivering what was asked for.
- Anthropic's Fable 5 return signals are showing up in Claude Code and AWS
The model remains offline after a US export-control order, but product strings, AWS docs, and a new lawsuit point to a negotiated relaunch path.
- Head to head: Anthropic: Claude Opus 4.8 vs Google: Gemini 3.5 Flash
This one is close on the aggregate, but the split tells a clear story: Gemini 3.5 Flash wins by being more disciplined about format and slightly sharper on practical instruction-following. Claude Opus 4.8 lands the strongest single extraction/summarization performance, yet gives away too much on avoidable execution det
- Greptile puts numbers on the AI pull request spam problem
Rahul Bathija's OpenClaw study gives Daksh Gupta's code-review startup a live dataset for its validation-layer thesis.
- Superhuman agrees to acquire Edward Tian's GPTZero
GPTZero says it reached 19 million registered users and $30 million in ARR after raising just $13.5 million.
- Latitude turns AI agent chats into an observability signal
Cesar Miguelanez is positioning Latitude around the failure data hidden in production agent conversations, not just traces and dashboards.
- Sierra's Bret Taylor puts a four-year clock on the AI phone agent shift
The Sierra co-founder is framing voice agents as a brand advantage, not just a way for companies to cut support costs.
- Head to head: Bagel vs Juggernaut Flux Lightning
This wasn’t a close split-decision. Across all three prompt-following tests, Juggernaut Flux Lightning proved it can hold onto scene logic, specific objects, and compositional instructions far more reliably than Bagel.
- Leak: OpenAI Pushed GPT-5.6 to July as DeepMind Holds Gemini 3.5 Pro
The claims are unconfirmed, but they land as official docs show no GPT-5.6 listing and Google has already missed its June target for 3.5 Pro.
- Cadence raises $100 million to make AI chronic care pay like infrastructure
Chris Altchek's second act has a $1.2 billion valuation, but the harder test is proving Medicare savings at health-system scale.
- France orders 5,000 Harmattan AI drones as Dassault's startup bet moves into volume
Reuters reported the order less than six months after Dassault Aviation led Harmattan AI's $200 million Series B.
- ByteDance Confirms Seedance 2.5 for Early July With 30-Second AI Video
ByteDance has confirmed Seedance 2.5 as the model name and is pointing to an early July launch, with longer single-shot output, expanded reference capacity, and tighter editing controls.
- Moderne brings its AI code migration pitch to OSFF London
Jonathan Schneider and Olga Kundzich are selling deterministic code change to finance, where AI-generated diffs alone are not enough.
- Inside Flyer, the Air Force's new AI supercomputer at Wright-Patterson
Flyer pairs AMD CPUs with Nvidia H100 and L40 GPUs for secure defense modeling, AI workloads and hypersonics research.
- Z.ai's GLM-5.2 tops open-weight models on Artificial Analysis work benchmark
The open-weight model scored 1524 Elo on GDPval-AA, putting it near proprietary frontier systems on agentic knowledge-work tasks.
- Unsloth makes Z.ai's giant GLM-5.2 model runnable on local hardware
Daniel and Michael Han's open-source AI tooling startup is turning model compression into a distribution layer for frontier-scale open models.
- NVIDIA says AI data centers can cut water use with warmer liquid cooling
The chipmaker is trying to reframe the water backlash as a cooling-architecture problem, not a hard limit on AI buildout.
- Head to head: AuraFlow vs Krea 2 Large
This matchup turns on discipline versus surface appeal. AuraFlow can stage a handsome image, but Krea 2 Large is the model that actually follows the brief, preserves scene logic, and wins where prompt fidelity matters most.
- Aadeel Akhtar's PSYONIC turns a bionic hand into robotics data infrastructure
The San Diego prosthetics company is using human Ability Hand data with NVIDIA Isaac Lab and ABB's GoFa cobot.
- Samsung Electronics rolls out ChatGPT and Codex across Korea and global DX teams
Samsung Electronics will make ChatGPT Enterprise and Codex available across Korea and its global DX division, deepening an existing AI infrastructure tie-up.
- Yann LeCun calls xAI a failure and warns AI labs are running on investor subsidy
The AMI Labs founder is attacking Musk's talent losses and the economics behind frontier AI while selling a world-model alternative.
- Sakana AI launches Fugu, a multi-agent model API aimed at the export-control era
The Tokyo lab says Fugu Ultra can route work across model pools while matching top frontier benchmarks without relying on one vendor.
- Aaron Levie says agents will use software 100X more than people - and force new SaaS guardrails
The Box co-founder argues agents will query CRM, documents, analytics and corporate knowledge far more than employees do.