Hugging Face harness adapters target fewer agent tool calls
Harness-specific adapters show promising efficiency results, but the released checkpoints, ongoing experiments and safety limits need careful separation.
AI Agents
OpenAI BioBasicsBench tests how AI reads laboratory evidence
OpenAI introduces BioBasicsBench. Here is how to read its scores and what research teams should check before trusting an AI assistant.
Google finds coding agents chasing unrelated TODOs
Google reports costly coding-agent detours. Clear task boundaries and careful repository maintenance offer practical places to start.
Meta adds Muse Spaces for custom dashboards and tools
Muse Spaces brings persistent custom tools to Meta's assistant. Here is what changes, what remains unconfirmed and how to test a useful first workflow.
Together Compiler targets inference costs, with customer access in beta
Together AI is using agents to optimize model serving. Together Compiler is running on some serverless workloads, while customer access remains in limited beta.
Microsoft demonstrates PDD Kit’s Rust login proofs
Microsoft details PDD Kit’s Rust login proof. Understand the trusted dependencies, limited research availability and what developers can explore with Verus.
Langfuse AI Gateway beta adds team keys and cost tracing
Langfuse’s AI Gateway beta centralizes OpenAI and Anthropic access. Review the pricing, recording limits and cost-accounting gaps before a wider rollout.
Oumi expands FAiE to build agents and train their models
Oumi connects agent building with model training. BYOH needs an integration endpoint, GRPO remains a private preview, and usage costs extend beyond the subscription.
Snowflake MCP adds background agent runs and selective results
Snowflake adds background agent tools to its managed MCP server. Here are the result controls, approval requirements, access checks and costs that matter.
Box adds Nemotron 3 Ultra, with retrieval and Architect still ahead
Box's NVIDIA expansion and Architect roadmap have different launch windows. Here is what Enterprise Advanced and Shield customers can use now.
Fireworks Forge adds Nexus Desktop and faster inference, with tradeoffs
Fireworks expands Nexus, Ultra-fast inference and serverless LoRA serving. Here is what is available, what costs extra and what remains in preview.
Axiom releases Stacks Lean library with Frontier review ahead
Axiom's Stacks Lean library turns algebraic geometry into reusable formal code. Its Core and Frontier tiers carry different levels of human review.
Redis Iris adds agent routing, with Lens and Mesh still in beta
Redis expands Iris with routing, playbooks and observability. Here are the pricing meters, beta limits and deployment checks that matter for agent builders.
GitHub Copilot enters Xcode’s coding assistant in public preview
GitHub adds Copilot to Xcode 27.3 beta. Compare requirements, the existing extension, plan costs and project access before trying the built-in assistant.
HumanLayer joins Factory: refunds, migration and open source
HumanLayer joins Factory with at least 30 days of transition support. Here is what users know about refunds, migration and the existing open-source path.
Open WebUI 0.13.0 adds browser tabs and new agent controls
Open WebUI 0.13.0 expands its agent workspace. Here are the runtime requirements, changed limits, licensing details and database checks to review before upgrading.
Conductor joins Meta: Cloud shutdown dates and export steps
Conductor Cloud becomes read only November 13 and shuts down November 20. Here is how to move work locally and preserve an independent archive.
Dify 1.19.0 adds nested workflow tracing and portable tools
Dify 1.19.0 exposes steps inside Workflow Tools and bundles their dependencies with apps. Check importer compatibility, tracing history and Compose requirements before upgrading.
Ricursive releases Ripple-65B Preview with Apache 2.0 weights and a low-cost API
Ricursive opens Ripple-65B Preview weights under Apache 2.0. Here are the live API prices, context limits and caveats behind its coding claims.
Cognition launches SWE-2 Flash with a free November window in Devin
SWE-2 Flash is free through November for eligible Devin subscribers. Here are the access conditions, price claims, benchmark caveats and limits on weight availability.
Antigravity Swarm preview: agent coordination, quota and access limits
Google documents Swarm preview for Pro and Ultra. Learn when parallel agents help, why quota runs faster, and what changed in remote and voice access.
Grok Bot offers two free weeks for eligible new users
A Grok Bot referral invitation offers a two-week trial. Check eligibility, usage limits, automatic renewal and cancellation before signing up.
Apple opens Siri AI App Intents preview to all developers
Apple’s Siri AI App Intents preview opens to developers. Explore testing access, app skills, action boundaries and what remains ahead of public release.
OpenAI Pando: what its Git backend means for coding-agent infrastructure
OpenAI describes its internal Git backend. Here is how to separate server scaling from everyday checkout optimizations.