Human researchers decide everything from compute and data spend to what experiments to run. To replace them, models need this judgement. We’re introducing FrontierML, a benchmark testing how well agents use money to scale model performance on 21 challenging, end-to-end ML tasks.

r-1 led K1Bench, Verona’s human-adjudicated benchmark of production document parsers, with 85.20% of pages parsed without a material error—9.2 percentage points ahead of the next-highest system. The benchmark used real-world K-1 packets supplied by a partner accounting firm. Reducto’s r-1 is built for exactly this kind of document, preserving both the content and the relationships that give it meaning.

Meet the new Subframe: built for designers who code. Most of your work is now in code. But when you need to explore or nudge the details, prompting isn’t enough. You need a canvas. Now moving between code and design is as simple as copy / paste.

We’re introducing intrinsic discovery, a new learning method that allows models to explore and learn from open-ended environments without human supervision. Using intrinsic discovery, we trained a world model that surpasses GPT-5.6 Sol at simulating terminal environments.

today, we're launching Light Parse Extend can now parse documents up to 70% cheaper than other solutions - light parse starts at $0.00625 per page, and scales down significantly with volume discounts - it's SOTA on Databricks OfficeQA Pro (a 3rd party agent benchmark) among all non-agentic parsing providers our ML/AI team trained smaller and more efficient layout, table, and form vision models to enable significant cost reduction without impacting accuracy on many types of documents live today for all users!

Big day for Ollama! When we started, open models and the open source AI ecosystem were in their early days with few believers. Our belief in open source has never wavered. With today's fundraising announcement and our 9M+ active builders, we’re ready to scale open models into AI that you can own. All aboard open models! 🧵

We've updated the Artificial Analysis Coding Agent Index, replacing SWE-Bench Pro with Datacurve's DeepSWE benchmark - the swap lifts Codex with GPT-5.5 (xhigh) above Claude Code with Opus 4.8 (max), while the newly released Claude Fable 5 (max) in Claude Code debuts at the top DeepSWE, built by @datacurve, writes its tasks from scratch rather than adapting them from public GitHub issues or pull requests, so no model has seen the solutions during training. That matters because SWE-Bench Pro, the benchmark it replaces in our Coding Agent Index, had grown gameable, with some models recovering the fix from the repository's commit history instead of solving the task. The swap reorders the index: Codex with GPT-5.5 (xhigh) rises from 65 to 76, overtaking Claude Code with Opus 4.8 (max) at 73. Claude Code with Fable 5 (max), which enters directly on the refreshed index, leads at 77. SWE-Bench Pro had been flattering some combinations and penalizing others. More below.

Today we’re releasing DeepSWE, a new standard for agentic coding benchmarks. On public leaderboards, top models often look relatively close in capability. DeepSWE shows where they actually diverge, reflecting the realistic experience of developers in their day-to-day work.

open sourcing an internal tool I built for me and @tcolitsas. its a browser agent that analyzes your codebase + logs -> figures out what UI flows are commonly traversed by users -> spins up cloud browsers + autonomously tests them hosted version: deployverona.com repo: github.com/verona-team/veron… hope its helpful for someone. built with @LangChain @agentmail @browserbase @modal @supabase @vercel

.@Aemon_ai is the AI R&D engineer. Give it a problem and a way to measure success, it then discovers optimal solutions, often going beyond what human experts can find. They just set a new record on circle packing — a classic NP-hard problem — with less than $10 of compute. Learn more at aemon.ai. Congrats on the launch, @Richard__Zhou and @RayXu_official! ycombinator.com/launches/PXs…