All Stories — Page 32

GPT-6.1 Sol Pro leads the equal-weight index, which combines three task suites with samples ranging from 50 to 100 tasks.

Anthropic’s Claude Opus 5.5 scored 1.00 on the 50-task sample at $0.0242 per task. OpenAI’s GPT-6 Astra and GPT-6.1 Sol Pro tied for second at 0.99, with Sol Pro costing $0.0121 per task.

Reflection AI CEO Misha Laskin is preparing its first open-weight model for customer-controlled AI systems, backed by compute agreements with SpaceX and Nebius and a planned South Korean data center.

Omnara's January 28th, 2026 post treats its remote-coding app as history and describes the managed-agent platform it now offers; the exact transition date remains unknown.

Upsolve AI's Ka Ling Wu and Serguei Balanovich carried lessons from Palantir's HyperAuto into a product that gives analytics agents company definitions, permissions and query checks.

Founders Karim Rahme and Tobias Herber are positioning Metorial as infrastructure for developers and enterprises, with identity, permissions and tracing alongside more than 1,000 integrations. Metorial's public evidence of demand remains early and largely company-reported.

The open-source project routes local training through a central coordinator; its creator is testing a consumer-hardware approach, not claiming a finished chatbot.

Hari Mulackal says Apex can make security-agent work cheaper; Cantina's own held-out test put it just behind Claude Opus 5 High at a fraction of the estimated cost.

Tibo Sottiaux replied to developer Ben Davis on October 4th; GPT-6.1 Sol launched on September 29th, and its promised faster mode has no firm release date.

An October 3rd roundup confirms the addition but names no models or terms. Framer's September 15th model guide lists seven choices, but predates the update.

ThinkingBox checks 507 simulated business workflows against their final records, with 20 runs per task to show the gap between a lucky success and repeatable work.

The October 3rd preview follows a May design lock for Figure's fourth-generation humanoid, but Figure has not published hand specifications or independent dexterity results.

A screen recording posted October 4th lists four paid tiers and shared AI usage; its author says the offer is still in testing.

The Anthropic engineer says he will bring the research-workflow approach he built for scientists to AI tools for doctors.

Founder Saoud Rizwan built Cline around developer choice; the offer also directs users to its standalone desktop beta.

Founder Ryan Merket reports a 36-minute median turnaround and 957 million tokens processed; traffic and revenue figures are still to come.

Cotool, founded by former Material Security operators, builds configurable agents for security teams. It raised a $7.4M seed led by a16z in March 2026.

Founded by Joseph Perrotta and Kaan Durmaz, Invertix says its agents cover about 2 GW of assets while human operators retain authority over physical actions.

Kolibri has 78.1 billion parameters, activates 3.46 billion per token and is released under Apache 2.0. Its model card calls for at least two A100 80GB GPUs.

Co-founder Henry Ndubuaku helped build Whistle into Cactus Compute's existing Needle runtime, adding keyword biasing and word timestamps for on-device applications.

An ISCA paper reports up to 100 TB/s per card, while its throughput comparisons remain modeled and Raptor is still pre-production.

Co-founder DJ Seo calls it arguably BCI's hardest problem, but the claim is not a published technical result.

Ai2 says AstaBrief averages 51.1 seconds per report in Asta Fast mode, versus 178.5 seconds for Claude-powered Thinking mode.

The Storyteller EPUB library modifies EPUBs for sentence-level audio alignment; v1 adds browser-based, read-only access and EPUB 2 upgrades.