Gemini 4 Argon compared with GPT-6 Astra, GPT-6.1 Sol and Claude Fable 5.1

Google describes Gemini 4 Argon as a frontier model for real-world coding, enterprise knowledge work and cyber defense, and said it would roll out soon.[1] Artificial Analysis reported that it was rolling out to selected users and was not publicly available.[2] The analysis also reported a 1-million-token context window.[2] For teams assessing it, those details make access a practical first checkpoint: the large context window does not mean the model is available to every prospective user.[2]

Artificial Analysis gave Argon an Intelligence Index score of 53, matching GPT-6 Astra and placing it one point above GPT-6.1 Sol.[2] It also ranked Argon first on AutomationBench-AA, with 78%, seven points ahead of Claude Sonnet 5.5 at maximum reasoning.[2] But Argon scored 50% on AA-Omniscience, five points below Gemini 3.1 Pro Preview and 13 points below GPT-6 Astra.[2] The results point to differences between evaluations, rather than a single score that establishes the best model for every task.[2]

Artificial Analysis reported standard API pricing of $4 per million input tokens and $20 per million output tokens for Argon, with an introductory 50% discount to $2 and $10.[2] It estimated an Intelligence Index task cost of $1.99 with the discount, or 60% of GPT-6 Astra’s $3.26.[2] After the discount, that Argon task cost would rise to $3.98, but Google had not confirmed when the promotion would end.[2] Teams comparing budgets should therefore distinguish the introductory estimate from the reported post-discount cost.[2]

OpenAI says GPT-6.1 Sol approaches GPT-6 Astra on agentic coding, computer use and professional work at one-fifth of Astra’s standard token prices; it also says Sol matches Astra at roughly one-fifth the cost.[3] Anthropic says Claude Fable 5.1 is generally available and estimated it would cost 25% less than Fable 5 for typical token-billed workloads.[5] By contrast, Anthropic’s Mythos 5.1 is limited to trusted-access programs.[5] OpenAI cautions that its factuality test used deliberately difficult prompts and is not representative of typical usage.[3]