New Company Agent Bench: Gemini Neon leads at 87.4%, simple API harnesses beat products
philhchen · x · 2026-11-21
philhchen's team released Company Agent Bench, a benchmark measuring how well company-assistant agent products understand their company, with first results:
- Gemini Neon takes a strong lead at 87.4%, ahead of Claude Fable 5.2 (82.2%) and GPT-6.2 Astra (79.7%).
- Even at low effort, Gemini Neon scores 85.3%, beating Fable 5.2, Opus 5.5, GPT-6.2 Astra and GPT-6.2 Sol at every reasoning effort tested.
- Notably, API models with a simple harness outperform the packaged products (Dots, Grok Bot, Claude Tag, Devin were compared).
- The author hints a new Google company-agent product may be coming.
More from coding & agent
- The sentences your customer-facing AI agent must never say—and how to test for them — rcleffke · 2026-11-21
- Haijun Xia pushes back on 'UI is dead': chat is a UI too, and far from sufficient — sh_reya · 2026-11-21
- Fully local pipeline: open-weight 27B models ship production code with Playwright + vision-loop tests — awitod · 2026-11-21
- Vespa co-founders launch Hornet: solving retrieval for the agentic era — jobergum · 2026-11-21
- Head-to-head coding test: Tencent Hy4 Flash finishes in 7:38 vs DeepSeek's 90 minutes — HeyToha · 2026-11-21
- Antigravity's new /teamwork feature lets you assign tasks to multiple agents at once — venturetwins · 2026-11-21