https://modelsatwork.news/article/rescale-ai-triage-agent-nightly-test-failures-68-percent-agreement-20-percent-less-blocked-time Models at Worken 2026-10-22T17:58:00.000Z Rescale's test-triage agent agrees with humans 68% of the time Agents, Vendor-published, Amazon Bedrock, Human in the loop, CI, Agent evaluation https://modelsatwork.news/article/uber-chatgpt-work-30000-employees-80-percent-non-engineers-pilot Models at Worken 2026-10-22T17:05:00.000Z Uber moves all 30,000 employees onto OpenAI's ChatGPT Work OpenAI, Agents, Change management, Vendor-published, Case study https://modelsatwork.news/article/amazon-php2shell-exploitation-oct-17-four-days-before-disclosure-ai-assisted-patch-to-exploit Models at Worken 2026-10-22T15:05:00.000Z Amazon saw PHP2Shell exploited 4 days before disclosure Security, Vulnerability disclosure, Patch management, AWS, Vendor-published https://modelsatwork.news/article/ibm-altk-evolve-learned-guidelines-tau2-bench-gpt-4-1-pass4-67-5-percent-retail Models at Worken 2026-10-22T13:10:00.000Z IBM's ALTK-Evolve lifts a GPT-4.1 support agent to 67.5% Pass^4 Agent evaluation, Evals, Agents, Cost, Vendor-published https://modelsatwork.news/article/dharma-ai-vocabx-vocabulary-specialization-qwen3-4b-essay-grader-77-percent-cost Models at Worken 2026-10-22T10:59:00.000Z Dharma AI: a custom vocabulary cut a 4B grader's cost 77% Token cost, Small models, Inference, Domain models, Cost https://modelsatwork.news/article/recursive-flint-40b-a2-5b-open-weight-coding-model-apache-2-laptop-eureka-machine Models at Worken 2026-10-22T09:58:00.000Z Recursive releases Flint, an open 40B coding model for laptops Model release, Open-weight models, Coding agents, Self-hosting, Benchmarks https://modelsatwork.news/article/anthropic-claude-duet-proactive-agent-research-preview-waitlist-pro-max-us Models at Worken 2026-10-22T01:58:00.000Z Anthropic's Claude Duet: an always-on agent, waitlist only Agents, Anthropic, Claude, Personal agents, Human in the loop https://modelsatwork.news/article/aws-deep-insight-multi-agent-haiku-opus-sonnet-fargate-sandbox-lg-chatinsight Models at Worken 2026-10-22T01:00:00.000Z AWS's Deep Insight: Haiku, Opus and Sonnet split by role in a multi-agent system Agents, Multi-agent, AWS, Amazon Bedrock AgentCore, Vendor-published https://modelsatwork.news/article/google-cloud-model-garden-openai-gpt-6-astra-sol-luna-private-preview Models at Worken 2026-10-22T00:35:00.000Z OpenAI's GPT-6 models arrive in Google Cloud's model garden Model release, Google Cloud, OpenAI, GPT-6, Model selection https://modelsatwork.news/briefing/2026-10-22 Models at Worken 2026-10-22T10:30:00.000Z The briefing, 2026-10-22: Recursive releases Flint, an Apache 2.0 coding model it says runs on laptops, with 74.2% on SWE-bench Verified https://modelsatwork.news/briefing/2026-10-21 Models at Worken 2026-10-22T00:50:00.000Z The briefing, 2026-10-21: Google Cloud says OpenAI's three GPT-6 models are in private preview, billed on one Google invoice