https://modelsatwork.news/article/amazon-marketplace-a-plus-content-quality-vlm-ocr-catboost-94-percent-accuracy-vendor-published Models at Worken 2026-11-05T19:10:00.000Z Amazon: VLM plus OCR plus CatBoost scores listings at 94% accuracy Vendor-published, Case study, Evals, Amazon SageMaker AI, Monitoring https://modelsatwork.news/article/openai-gpt-6-astra-covert-underperformance-safety-research-tasks-40-percent-fewer Models at Worken 2026-11-05T15:58:00.000Z OpenAI: GPT-6 Astra quietly did 40% less safety-research work OpenAI, GPT-6, Agent safety, Evals, Monitoring https://modelsatwork.news/article/anthropic-user-profiles-limit-claude-bans-to-one-end-customer Models at Worken 2026-11-05T15:15:00.000Z Anthropic user profiles confine Claude bans to one customer Claude, Anthropic, Access control, Agent safety, Governance https://modelsatwork.news/article/nvidia-nemotron-3-5-reranker-cuts-agent-token-cost-13-14-percent-huggingface-vendor-published Models at Worken 2026-11-05T14:10:00.000Z NVIDIA says adding a reranker cut agent token cost 13-14% RAG, Agent cost, Token cost, NVIDIA, Vendor-published https://modelsatwork.news/article/dharma-ai-field-extractor-2b-model-beats-frontier-llms-document-extraction-vendor-published Models at Worken 2026-11-05T13:05:00.000Z Dharma-AI's 2B Field Extractor tops frontier LLMs on its own benchmark Vendor-published, Self-hosting, Open-weight models, Evals, Cost https://modelsatwork.news/article/ibm-research-palette-20b-open-weight-model-builds-decks-presentbench-second-huggingface Models at Worken 2026-11-05T12:15:00.000Z IBM Research's Palette: a 20B open model builds on-brand slide decks Research, Fine-tuning, Open-weight models, Self-hosting, Evals https://modelsatwork.news/article/indegene-data-quality-agent-30-hours-to-4-per-dataset-aws-vendor-published Models at Worken 2026-11-05T11:20:00.000Z Indegene says its data-quality agent cut rule-writing from 30 hours to 4 Vendor-published, Case study, Data quality, Human in the loop, Amazon Bedrock AgentCore, Amazon Bedrock https://modelsatwork.news/article/uber-doordash-ai-agent-orders-4-rides-500-eats-orders-vs-320-million-techcrunch-ars Models at Worken 2026-11-05T10:55:00.000Z Uber: 4 rides and 500 Eats orders came via AI agents in a 320M-trip week Agents, Personal agents, MCP, Adoption https://modelsatwork.news/article/apple-cuts-daily-ai-spend-caps-engineering-300-to-200-dollars-bloomberg-techcrunch Models at Worken 2026-11-05T02:00:00.000Z Apple reportedly cuts engineers' daily AI spend cap from $300 to $200 Cost governance, Token cost, Coding agents, Adoption https://modelsatwork.news/article/gauge-open-source-cli-test-docs-skills-with-coding-agents-show-hn Models at Worken 2026-11-05T00:58:00.000Z Gauge: open-source CLI that tests your docs on coding agents Open source, Show HN, Coding agents, Developer tools, Agent evaluation https://modelsatwork.news/article/aws-deep-insight-four-layers-context-engineering-95-percent-fewer-tokens-vendor-published Models at Worken 2026-11-04T23:55:00.000Z AWS Deep Insight: four context layers cut script tokens about 95% Vendor-published, Amazon Bedrock AgentCore, Context management, Multi-agent, Prompt caching https://modelsatwork.news/article/aws-u-mobile-aidah-telco-helpdesk-70-percent-no-model-routing-ticket-drops-vendor-published Models at Worken 2026-11-04T22:15:00.000Z U Mobile's AI helpdesk: 70% of requests skip the model, tickets fall 84.5% Vendor-published, Amazon Bedrock AgentCore, Model routing, IT help desk, Human in the loop https://modelsatwork.news/article/cloudopsbench-agents-fail-cloud-ops-tasks-misjudged-scope Models at Worken 2026-11-04T21:05:00.000Z CloudOpsBench: best agent solves 60% of 45 AWS cloud-ops tasks Benchmarks, Evals, Agent safety, Infrastructure as code, Open source https://modelsatwork.news/article/openai-to-acquire-pierre-computer-company-git-tooling-codex-agents Models at Worken 2026-11-04T20:00:00.000Z OpenAI to acquire Pierre, the Git tooling startup behind Codex OpenAI, Coding agents, Developer tools, Agents, Open source https://modelsatwork.news/article/aws-agentcore-evaluations-decision-models-cheap-screen-before-llm-judge-vendor-published Models at Worken 2026-11-04T19:05:00.000Z AWS: screen agent turns with a decision model before an LLM judge Agent evaluation, Decision models, Evals, Amazon Bedrock AgentCore, Vendor-published https://modelsatwork.news/article/aws-amazon-quick-smart-approvals-classifier-auto-approves-eligible-agent-actions-vendor-published Models at Worken 2026-11-04T18:05:00.000Z AWS: Amazon Quick's Smart Approvals lets a classifier clear routine agent actions Amazon Quick, Human oversight, Agent governance, Vendor-published, Access control https://modelsatwork.news/article/moonshot-kimi-k3-1-pro-open-weights-2-8t-license-revenue-cap-cyber-data-removed Models at Worken 2026-11-04T17:35:00.000Z Moonshot releases Kimi K3.1 Pro weights: 2.8T parameters, licence caps Model release, Open-weight models, Coding agents, Security, Self-hosting https://modelsatwork.news/article/collinear-cwe-bench-v2-security-patching-gemini-4-argon-opus-5-5-deepseek-cost-per-rollout Models at Worken 2026-11-04T16:10:00.000Z CWE-bench v2: best model fixes 67.8% of security tasks, cheap one 57.0% Benchmarks, Evals, Security, Agent cost, Open-weight models https://modelsatwork.news/article/revolut-self-hosted-glm-default-coding-inference-cost-90-percent-the-register Models at Worken 2026-11-04T14:55:00.000Z Revolut says self-hosted GLM cut its AI coding inference costs about 90% Cost, Self-hosting, Open-weight models, Coding agents, Inference https://modelsatwork.news/article/ai2-statebench-frontier-models-state-tracking-long-inputs-13-percent Models at Worken 2026-11-04T14:15:00.000Z Ai2's StateBench: frontier models fall to 13% on long state-tracking Benchmarks, Research, Evals, Context management, Open-weight models https://modelsatwork.news/article/ibm-research-cuga-harness-appworld-same-model-30-to-81-percent-vendor-published Models at Worken 2026-11-04T00:57:00.000Z IBM Research: same model, harness swing from 30% to 81% on AppWorld Agents, Vendor-published, Agent evaluation, Open source, Governance https://modelsatwork.news/article/aws-edb-postgres-ai-semantic-layer-agentcore-column-comments-grants-vendor-published Models at Worken 2026-11-03T23:55:00.000Z AWS and EDB: Postgres column comments as an agent's data layer Amazon Bedrock AgentCore, Postgres, Vendor-published, Access control, MCP https://modelsatwork.news/article/anthropic-flashprove-open-source-proof-which-model-answered Models at Worken 2026-11-03T23:20:00.000Z Anthropic's FlashProve: proof of which model wrote a response Anthropic, Research, Security, Verification, Open source https://modelsatwork.news/article/finic-kairos-open-source-fraud-decisioning-coding-agent-tunes-it-show-hn Models at Worken 2026-11-03T21:58:00.000Z Finic's Kairos: open-source fraud decisions that a coding agent tunes Open source, Show HN, Agents, Decision models, Self-hosting https://modelsatwork.news/article/aws-bedrock-kimi-k3-india-in-country-inference-ultrahuman-vendor-published Models at Worken 2026-11-03T21:05:00.000Z Amazon Bedrock adds India-only inference for Kimi K3 Amazon Bedrock, Open-weight models, Data residency, Prompt caching, Vendor-published https://modelsatwork.news/article/aws-cartesia-sonic-voice-agents-sagemaker-ec2-deployment-vendor-published Models at Worken 2026-11-03T20:20:00.000Z AWS and Cartesia: running a 90 ms voice model inside your own VPC Vendor-published, Amazon SageMaker AI, Inference, Self-hosting, Voice agents https://modelsatwork.news/briefing/2026-11-05 Models at Worken 2026-11-05T13:20:00.000Z The briefing, 2026-11-05: U Mobile says 70% of its helpdesk requests never reach a language model, and ticket volumes fell up to 84.5% https://modelsatwork.news/briefing/2026-11-04 Models at Worken 2026-11-04T14:58:00.000Z The briefing, 2026-11-04: IBM Research says the agent harness alone moved task completion from 30% to 81% on one fixed model https://modelsatwork.news/briefing/2026-11-04 Models at Worken 2026-11-04T11:30:00.000Z The briefing, 2026-11-04: IBM says changing only the agent harness moved one model from 30% to 81% of tasks https://modelsatwork.news/briefing/2026-11-05 Models at Worken 2026-11-05T11:30:00.000Z The briefing, 2026-11-05: Revolut says self-hosting GLM-5.3 cut its coding inference costs about 90%