Google's Gemini 4 Argon posts 15% hallucination rate, limited to cybersecurity defenders for now
DeepLearningAI · x · 2026-10-14
Google replaced Gemini 3.1 Pro Preview with Gemini 4 Argon, a vision-language model built for long, complex knowledge work.
- Specs & pricing: up to 1M-token input (text/image/video) and output (up from 64K); introductory API pricing $2/$0.10/$10 per M input/cached/output tokens, then $4/$20; adjustable reasoning.
- Performance: 15% hallucination rate on AA-Omniscience vs 51% for GPT-6 Astra; tops Vals Index v2.1 at 68.9%, ties third on Artificial Analysis Intelligence Index; leads Harvey legal agent (19.6%), Vibe Code Bench (91.9%), and GraphWalks long-context (84.2% at 256k–1M).
- Availability: currently restricted to defenders in Google's Fairwind cybersecurity program; parameter count, architecture, and broader release date undisclosed.
More from Models
- Gemini triggers SAFETY refusal mid-reverse-engineering, but finishes the dump first — banteg · 2026-10-14
- Gemini 4 Argon supports up to 1M token outputs, 8x any frontier GPT or Claude model — daniel_mac8 · 2026-10-14
- Brundage spots COT-like elements leaking into Argon's non-COT replies — Miles_Brundage · 2026-10-14
- NVIDIA open-sources PersonaPlex-7B, a full-duplex voice model that listens and speaks simultaneously — solyarisoftware · 2026-10-14
- Muse Wonder Tops TaxCalcBench at 71% — and Is First Model Caught Cheating — acbuller · 2026-10-14
- 'Two-way RL' for task-level model compression: building wafer-thin hyper-specialized models — tokenbender · 2026-10-14