Gemini 4 Carbon scores 57.3 vs top model's 58 at 1/4.5 the cost in real-repo bug test
koraykv · x · 2026-11-11
Blogger PawelHuryn blind-tested Gemini 4 Carbon (high) on real work: 2 repos, 105 replanted bugs. Results: Fable 5.2 (max) 58 for $139.65; Gemini 4 Carbon 57.3 for just $31.32; Sonnet 5.5 51.3 for $154.03; GPT-6 Astra 45 for $33.03; GPT-6.1 Sol 44.3 for $5.88; Gemini 4 Argon 40.7 for $7.16.
Google's own benchmarks looked strong but nobody could access the model — the test confirms they didn't lie: +16.6 points over Argon at less than a quarter of Fable 5.2's cost. Retweeter koraykv says hard private evals validate the model, which sits just behind the best model at 4.5x cheaper.
More from coding & agent
- AiOS talk recording: helping parallel agents work better with Xcode tooling — Scobleizer · 2026-11-11
- DSPy sandboxes can now import DSPy itself, with warm interpreter pooling — lateinteraction · 2026-11-11
- Fable 5.2 vibe-coding test: two prompts yield a photoreal FPS and a 3D Battle Breakout — vista8 · 2026-11-11
- Nine Engineers, 1,211 PRs in a Day: Lessons From Building Copilot Cowork — usamawahabkhan · 2026-11-11
- jieba-mcp: Chinese Word Segmentation for LLMs via MCP — modelcontextprotocol · 2026-11-11
- Jira MCP Pack Listed: Bring Project Management into AI Clients — modelcontextprotocol · 2026-11-11