Gemini 4 Argon ties GPT-6 Astra on intelligence

Google DeepMind's Gemini 4 Argon, with high reasoning enabled, scores 53 on the Artificial Analysis Intelligence Index. That ties GPT-6 Astra (max) and puts it one point ahead of GPT-6.1 Sol (max, 52), according to independent benchmarking by Artificial Analysis. Argon is Google's first proprietary model above the Flash class in more than seven months. It scores 23 points higher than Gemini 3.1 Pro Preview and 12 higher than Gemini 3.8 Flash (high).

Artificial Analysis says most of the gain comes from fewer hallucinations and better agentic performance. On AA-Omniscience, Argon's hallucination rate is 15%, the lowest of any model scoring 45 or more on the index. GPT-6 Astra comes in at 51% and GPT-6.1 Sol at 54%. Argon's accuracy is lower, though: 50%, against 63% for Astra. Argon ranks first on AutomationBench-AA at 78% and scores 57% on Terminal Bench 4. On Terminal Bench 4 it trails Claude Sonnet 5.5, Claude Opus 5.5 and GPT-6 Astra.

The cost picture rests on a launch promotion. Standard pricing is $4/$20 per million input/output tokens, currently cut 50% to $2/$10, and cached input is discounted 95%. At the promotional price, an Intelligence Index task costs $1.99, which is 60% of GPT-6 Astra's cost and 2.7x GPT-6.1 Sol's. At standard pricing it rises to $3.98, about 1.2x Astra. Argon is not cheaper because it uses fewer tokens: it averages 62k output tokens per task, compared with 27k for Astra.

The model has a 1M-token context window and accepts text, image, video and speech input. Gemini 4 Argon is currently available only to selected users. Google has not said when the discount ends.