https://modelsatwork.news/article/amazon-bedrock-structured-outputs-claude-json-schema-strict-tool-use-limits Models at Worken 2026-10-20T09:58:00.000Z Amazon Bedrock adds schema-enforced JSON for Claude, with limits Amazon Bedrock, AWS, Claude, Vendor-published, Tool use https://modelsatwork.news/article/microsoft-guardian-agents-agentic-insider-60-local-agents-dynamic-permissions Models at Worken 2026-10-20T01:05:00.000Z Microsoft adds Guardian Agents to treat AI agents as insiders Microsoft, Agent security, Access control, Agent governance, Monitoring https://modelsatwork.news/article/torchneuron-pytorch-backend-aws-trainium-hugging-face-89-percent-architectures Models at Worken 2026-10-19T22:05:00.000Z AWS TorchNeuron brings native PyTorch to Trainium; public beta 20 Oct Inference, AWS, Developer tools, Open source, Vendor-published https://modelsatwork.news/article/aws-vllm-push-based-kv-transfer-3x-faster-first-token-disaggregated-inference Models at Worken 2026-10-19T21:05:00.000Z AWS: vLLM push-based KV transfer cuts first-token time up to 3x Inference, Self-hosting, AWS, Vendor-published https://modelsatwork.news/article/twelvelabs-pegasus-1-5-amazon-bedrock-video-to-timestamped-json-two-hour-limit Models at Worken 2026-10-19T18:50:00.000Z TwelveLabs Pegasus 1.5 on Amazon Bedrock turns video into timestamped JSON Model release, Amazon Bedrock, AWS, Vendor-published, Video AI https://modelsatwork.news/article/aws-vllm-bi-directional-kv-transfer-agent-first-token-latency-1-6x-3-3x Models at Worken 2026-10-19T17:56:00.000Z AWS: vLLM bi-directional KV transfer cuts agent first-token time 1.6-3.3x Inference, Self-hosting, AWS, Agents, Vendor-published https://modelsatwork.news/article/deepgram-nova-3-aws-trainium2-11-70-percent-more-audio-per-dollar Models at Worken 2026-10-19T17:10:00.000Z Deepgram ports Nova-3 to AWS Trainium2: 11-70% more audio per dollar Inference, AWS, Cost, Vendor-published, Case study https://modelsatwork.news/article/claude-opus-5-5-fast-mode-open-to-all-paid-tiers-2x-price Models at Worken 2026-10-19T16:10:00.000Z Claude Opus 5.5 fast mode is open to all paid tiers, at twice the price Anthropic, Claude, Pricing, Inference, Cost https://modelsatwork.news/article/hugging-face-llama-windows-tray-app-local-models-openai-compatible Models at Worken 2026-10-19T15:00:00.000Z Hugging Face's Llama Windows runs open models locally, no admin rights Self-hosting, Open source, Open-weight models, Developer tools, NVIDIA https://modelsatwork.news/article/whatsapp-mcp-cloudflare-read-only-agent-access-baileys-terms-risk Models at Worken 2026-10-19T11:58:00.000Z whatsapp-mcp: a read-only WhatsApp server for agents, on Cloudflare Open source, Show HN, Cloudflare Workers, Personal agents, Security https://modelsatwork.news/article/google-research-harness2-agent-harness-self-refinement-benchmarks Models at Worken 2026-10-19T13:56:00.000Z Google Research's Harness²: agents that rewrite their own harness Research, Agents, Open source, Context management, Benchmarks https://modelsatwork.news/article/who-afro-claude-skill-ebola-situation-report-under-an-hour-anthropic Models at Worken 2026-10-19T13:15:00.000Z WHO AFRO uses a Claude skill to cut Ebola sitreps to under an hour Case study, Vendor-published, Healthcare, Claude, Human in the loop https://modelsatwork.news/article/strot-compiles-llm-intent-once-runs-144m-model-locally-60ms-per-record Models at Worken 2026-10-19T09:58:00.000Z STROT: use an LLM once, then decide locally at 60 ms a record Open source, Show HN, Decision models, Routing, Cost https://modelsatwork.news/article/liquid-ai-lfm2-5-24b-a2b-open-weight-on-device-agents-vendor-benchmarks Models at Worken 2026-10-19T02:10:00.000Z Liquid AI releases LFM2.5-24B-A2B, open-weight model for on-device agents Model release, Open-weight models, Agents, Inference, Vendor-published https://modelsatwork.news/article/anthropic-commits-150-million-over-three-years-to-genesis-mission-claude-credits Models at Worken 2026-10-19T00:05:00.000Z Anthropic commits $150m over three years to the Genesis Mission Anthropic, Claude, Vendor-published, Governance https://modelsatwork.news/article/deslop-oxlint-rules-reject-ai-generated-typescript-21-default-rules-mit Models at Worken 2026-10-18T23:10:00.000Z Deslop: Oxlint rules that flag AI-written TypeScript patterns Open source, Developer tools, Coding agents, Show HN, Code review https://modelsatwork.news/article/aachen-court-google-ai-mode-own-statements-liable-doctor-250000-euro-fine Models at Worken 2026-10-18T22:05:00.000Z Aachen court: Google's AI Mode answers are Google's own statements Governance, Regulation, Google, Liability, Hallucinations https://modelsatwork.news/article/hybridagent-latent-reasoning-cuts-agent-tokens-35-to-55-percent Models at Worken 2026-10-18T21:00:00.000Z HybridAgent paper: 35-55% fewer tokens by thinking in latent space Research, Agents, Inference, Token cost, Evals https://modelsatwork.news/article/provenanceguard-checks-agent-answers-name-right-mcp-source Models at Worken 2026-10-18T19:58:00.000Z ProvenanceGuard catches 138 of 139 wrong-source claims in agent answers Agents, Evals, Verification, RAG, Healthcare https://modelsatwork.news/article/deepmind-four-gemini-4-argon-agents-beat-minecraft-27-hours-1620-dollars-single-run Models at Worken 2026-10-18T18:05:00.000Z DeepMind: four Gemini 4 Argon agents beat Minecraft in 27 hours Agents, Multi-agent, Google, Agent cost, Research https://modelsatwork.news/article/anthropic-advanced-ai-framework-frontier-developers-10-25-flop-500m-revenue Models at Worken 2026-10-18T17:00:00.000Z Anthropic's policy proposal would regulate only the largest AI developers Governance, Regulation, Anthropic, Safeguards https://modelsatwork.news/article/websites-block-ai-agents-yelp-delta-jpmorgan-schwab-amazon Models at Worken 2026-10-18T15:58:35.000Z Yelp, Delta, JPMorgan and Schwab are blocking consumer AI agents Agents, Browser agents, Access control, Personal agents, Agent governance https://modelsatwork.news/briefing/2026-10-20 Models at Worken 2026-10-20T10:30:00.000Z The briefing, 2026-10-20: Anthropic says Opus 5.5 fast mode is open to all paid tiers at twice the price for up to 2.5x faster output https://modelsatwork.news/briefing/2026-10-19 Models at Worken 2026-10-19T10:30:00.000Z The briefing, 2026-10-19: Vals AI says a refund agent told to "try everything" sent 187 emails and shared a user's family illness