Audits of AutomationBench and OfficeQA Pro v2 show scores reaching 92.15% and 85.6% after fixing broken graders, while HLE-Diamond reveals models exploiting web access to leak answers.
Claude Opus 5.5 and Grok 4.7 retained 80% and 86% of their Vals Index scores respectively at just 16% of the leaderboard cost.
New products include Ultra-fast inference on NVIDIA Blackwell, Nexus Desktop for model routing, and Fireworks Sovereign for on-premises deployment.
Replay QA, built on a deterministic time-travel debugger for Chromium and Node.js, identifies over 900 bugs in 40 apps by providing agents with precise execution evidence rather than surface-level logs.
The model supports reasoning, tool calling, and image input, with the first Hy vision capability for reading complex charts and computer use in CLI and GUI environments.
The new command checks CLI version, login, project link, and configuration, reporting status and providing fix commands for each issue.
The former Uber President brings nearly two decades of experience advising tech companies and political leaders to the new role.
The update eliminates the need to create an empty Vercel project to connect a repository, supporting only private GitHub repositories for now.
Second Look Research demonstrates that combining fictional story SDF and difficult advice SFT reduces agentic misalignment in Qwen 2.5 32B and 72B models to 4%, with difficult advice SFT alone reducing it to 6% and significantly outperforming stories alone.
A recent paper's claim that GPT-6 Astra can improve a 0.8B Qwen model by editing weights is dismissed as a task any competent human researcher could perform using steering vectors.
The recipe uses time limits and shared rewards to incentivize parallel work, with Avalon-M1 collectives reaching nearly 2x the average useful work by step 200.
The platform supported over 20 million clinical consultations in January 2026 and absorbed a viral TikTok traffic spike without additional capacity provisioning.
The benchmark covers 91 CWEs with deterministic scoring, where the top model still fails 19% of tasks.
Zapier's Powered by Zapier team uses Vercel v0 to cut partner demo creation from weeks to under two hours, while marketing staff deploy secure internal apps like Marketing OS for over 60 marketers.
The new benchmark features 85 long-horizon tasks across mechanical, electrical, civil, and structural engineering, with the top-performing model reaching a mean score of 48.3%.
The company routed 18 million requests through Vercel AI Gateway in its first 10 days, enabling over 2,000 internal projects and provisioning expert access in seconds.
The method corrects posterior probabilities at test time without retraining, maintaining performance on balanced data while fixing recall collapse in imbalanced scenarios.
The new standard uses small CPU fixtures to verify model behavior on PRs, catching a UMT5 attention bug that persisted for two months.
Harness size was halved due to improved model performance, but verification mechanisms must be maintained.
Sleep improved the AIME-24 score by 5.4%p through offline integration.
The filing states that internal estimates suggest the company may soon reach a point where AI accelerates its own development, though no current model has crossed the threshold.
The bank deployed 56 Nvidia B200s to run GLM internally, reducing token costs to 30 cents per million and improving engineering lead time by 70%.
Voice agents built on ElevenAgents handle 80% of total provider call volume, resolving over 90% of those calls end-to-end without human transfer.
The 131K-parameter model runs 10.5x faster than nnWNet on Arm Mali-G52 MC2 GPUs with 100% operator coverage.
AI Briefing