"Nobody I know uses Gemini for coding" sparks debate as a sweeping Gemini benchmark chart circulates
vedantmisra · x · 2026-11-20
A resonant tweet observes that not a single person the author knows uses Gemini for coding, asking why Google—having scoured decades of web-scale code—still can't win developers over.
The attached chart shows "Gemini 4 Neon" sweeping Claude Opus 5.5 and GPT-6.2 Sol across benchmarks (CursorBench 4.1 at 65.4%, SWE-sweep provider harness 39%, GraphWalks long-context at 95.2%), but the model names are unreleased and the chart's authenticity is doubtful—treat it as community speculation rather than confirmed data.
The real tension: strong benchmark scores don't equal developer adoption; workflow lock-in around Claude Code, Codex, and Cursor may matter more than leaderboards.
More from Models
- Cost chart: Claude Fable 5.2 burns $54 per task, OpenAI a fraction — haider1 · 2026-11-20
- Stripe launches coding-agent integration evals; GPT-6-Astra tops one-time payments at 90 — jeff_weinstein · 2026-11-20
- Basalt 0.5.0 runs Tencent Hy4 Flash locally at 172 tok/s on a 5090 with 1M context — jesdga95 · 2026-11-20
- Matthew Berman teases Google's new 'Neon' model: 'very good' — MatthewBerman · 2026-11-20
- Dev runs 6 parallel Neon agents; benched Claude for too many safety refusals — iskander · 2026-11-20
- Granola switches default model to GPT-6 Sol, criticized as dumber and 2x pricier than GPT-6.1 — gabriel1 · 2026-11-20