Gemini 4 breakdown: Neon is 2x faster than Carbon but hallucinates more
teortaxesTex · x · 2026-11-19
teortaxesTex analyzes Artificial Analysis charts comparing Google's Gemini 4 Neon and Carbon variants:
- Non-Hallucination Rate: Carbon tops at 96%, Carbon (high) 89%, Neon (high) 86%; Claude Opus 5.5 (max) 65%, while Anthropic's Astra/Fable lines sit at 38-50%.
- Terminal-Bench-Science 0.1 (agentic scientific workflows): OpenAI Astra (xhigh) leads at 79%, Neon (max) 70%, Carbon 62-66%.
- His take: Neon matches or beats Carbon on AA-Accuracy but loses on non-hallucination — a smaller-model/lower-precision smell. Neon is 2x faster, more verbose, marginally cheaper per token: a classic Flash profile, though not billed as a separate pretrain; "MRL is just a handwave".
Overall: Neon's intelligence is legit; it's a speed-vs-hallucination tradeoff.
More from Models
- Gemini 4 Neon (High) beats Hydrogen and Carbon on AI SEO advice in hands-on test — gaganghotra_ · 2026-11-19
- Agents spent years hillclimbmaxxing while the human at the keyboard became an afterthought — tokenbender · 2026-11-19
- Speculation: Argon/Carbon/Neon may be nested distillations, with 'adapted' dodging 'post-trained' — teortaxesTex · 2026-11-19
- 6.2-Astra reportedly crushes contest benchmarks, finding a novel AWTF26 solution in 2 hours — i_dg23 · 2026-11-19
- KVARK Flash: a proxy that caps DeepSeek's thinking per request, 30-50% fewer tokens — Ill_Contribution_717 · 2026-11-19
- FoodTruck Bench, an LLM agent business-sim benchmark, lands on Papers with Code — NielsRogge · 2026-11-19