ARES is built around three pillars: 1. A familiar environment loop, modern concurrency 2. The correct RL training boundary 3. Integration with the OSS task ecosystem ( Harbor Framework )

ARES uses the Harbor task format ( Alex Shaw ). It comes with SWE-Bench Verified, TerminalBench2, SWESmith, and everything else in the Harbor ecosystem. We're also releasing 1k new JavaScript tasks with Vmax ( Augustine Mavor-Parker Matthew Sargent ) to help the ecosystem grow.

Alex Shaw Gotta love free datasets! Harbor is pretty slick.

SkyRL 🤝Ares🤝 Harbor

Joan Cabezas Martian Harbor Framework It's a good framework

Opus 4.6 is state-of-the-art on several evaluations including agentic coding, multi-discipline reasoning, knowledge work, and agentic search. We're also shipping new features across Claude in Excel, Claude in PowerPoint, Claude Code, and our API to let Opus 4.6 do even more.

Exciting mention of TBench 2.0 in today's model releases - congrats to Mike A. Merrill Alex Shaw & team + proud of Snorkel AI 's contributions! Benchmarks are just one (limited) measurement tool - but critical guideposts of frontier progress. Much more to build here ahead!

Terminal-Bench 2.0 (built on Harbor) getting lots of love today!

More Harbor format RL training environments! Now on the Harbor registry: harborframework.com/registry/seta-…

TerminalBench 2.0 & OSWorld benchmarks are having their moment - they seem to be one of the main focuses of the model benchmark reports, and I'm glad to finally see more focus on what models can do overall on a computer (beyond coding). We haven't even scratched the surface of

It is our other eval leveraging the great Harbor Framework by Alex Shaw Ryan Marten from Laude Institute .

Our ability to measure AI has been outpaced by our ability to develop it. This evaluation gap is one of the most critical problems in AI. Today, we’re excited to announce the Open Benchmarks Grants - with a starting $3M commitment from @Snorkel + support from @HuggingFace

We are partnering with Snorkel AI on Open Benchmarks Grants. This is an amazing opportunity to build the next generation of great evals. Come build your benchmark with Harbor and Snorkel!

Building Better Coding Agent Harnesses at LangChain we're thinking hard about the science of harness engineering + open research on what works & doesn't A quick peak on our deepagents X Terminal Bench 2.0 work, shoutout to Alex Shaw & Harbor (they're great). Broad research

Huge props to Alex Shaw and the folks at terminalbench / Harbor Framework — without Harbor, CCBench would’ve taken us months to ship instead a week. We actually tried this last year and gave up. Only decided to give it another shot because Harbor was released.

Ship benchmarks in weeks with Harbor. Congrats Paul Kuruvilla and team!

Your recurring reminder that Harbor has lots of datasets. Integrate once and get them all for free. harborframework.com/registry

SkyRL now implements the Tinker API. Now, training scripts written for Tinker can run on your own GPUs with zero code changes using SkyRL's FSDP2, Megatron, and vLLM backends. Blog: novasky-ai.notion.site/skyrl-tinker 🧵

Agent Skills are everywhere - Claude Code, Gemini CLI, Codex all support them. But do they actually work? 105 domain experts from Stanford, CMU, Berkeley, Oxford, Amazon, ByteDance & more built SkillsBench to find that out. 86 tasks. 11 domains. 7,308 trajectories. 🧵👇