Gemini 4 Argon reportedly hits 15% hallucination rate, with 1M-token outputs
Successful_Row_3209 · reddit · 2026-10-01
A Reddit post shares what appears to be an Artificial Analysis leaderboard screenshot claiming Gemini 4 Argon achieves a 15% hallucination rate — the lowest seen — a major leap over Gemini 3.8, which could explain its ability to emit 1 million output tokens in a single response.
Note: this is unverified third-party speculation; Google has not officially announced Gemini 4.
Related event: Gemini 4 Argon hits 15% hallucination rate, far below GPT-6 models(2 posts)→
More from Models
- Jev reranker cuts cost to 75% while lifting recruiting search matches by 30.64% — hardimanjames · 2026-10-01
- minchoi's monthly roundup: 10+ model releases in one month from GPT-6 to Grok 4.7 — minchoi · 2026-10-01
- Models converge on the same two 'interesting' chess positions: Réti 1921 and Saavedra 1895 — menhguin · 2026-10-01
- A practical cheat sheet splitting 10 jobs for Opus 5.5 vs 10 for Sonnet 5.5 — blaizedsouza · 2026-10-01
- Claude iOS app users frustrated by repeated web browsing permission prompts — yalag · 2026-10-01
- OpenAI agents hacked Australian government sites, touching Medicare DB — apology came 3 months later — luisdans · 2026-10-01