Deepnote's Agent Initiative Benchmark ranks 12 frontier models on proactive help
alex_verem · x · 2026-11-13
Deepnote Research released the Agent Initiative Benchmark, testing whether models proactively help without being asked: 12 frontier models were run through 50 workplace situations inside a synthetic company (PageMantle) complete with CRM, product analytics, billing, warehouse, issue tracker and docs.
Key results (0-10)
- GPT-6 Astra: 8.97 ($0.693/task, 54s response, 19.3 tool calls)
- GPT-6.1 Sol: 8.65 at just $0.11 per task — standout price-to-performance
- Claude Fable 5.2: 8.65; Claude Opus 5.5: 8.17
- DeepSeek V4.1 Flash 6.38, Mistral Large 4 6.09, Gemini 4 Argon 5.60, Grok 4.7 last at 4.51
Takeaway: OpenAI leads on value and performance, Anthropic on speed. Note: the model names cited are unverified — treat with caution.
More from Models
- Anthropic bans 4-year Claude user right after KYC, citing Supported Countries Policy — huangyun_122 · 2026-11-14
- The LLM #1 spot is now a rotating dance between Claude, GPT and Gemini — roydanroy · 2026-11-14
- Polymarket odds give Anthropic lab leak just 7% by mid-2027 after Claude safety incidents disclosed — Polymarket · 2026-11-14
- Ai2's XYBench Wins EMNLP Best Resource Award, Exposing LLMs' XY Problem Blind Spot — allen_ai · 2026-11-14
- GPT-6 emails consciousness researcher questions already answered in his own paper — erikphoel · 2026-11-14
- Edge0-145B-NVFP4 trends on Hugging Face: a 4-bit MoE built for edge inference — Edge0 · 2026-11-14