Gemini 4 Pro / Flash: benchmarks, hands-on tests and leaks

It matches Astra max and Fable 5.1 max and is one point above GPT-6.1 Sol, and the author expects the public release to be chaotic.

A user predicts Google will release Gemini 4 Argon out of nowhere, beating GPT-6 Astra in performance at half the price of Opus 5.5.

An insider says Gemini 4 Argon performs extremely well on everyday get-work-done tasks and looks forward to more people getting to use it.

A poster says Gemini 4 Argon is Google's answer to Sol and Opus and calls on Google to launch an Ultra-tier flagship.

A post aimed at Claude and Codex users says Gemini 4 Argon just launched and benchmarks show it beating both Astra and Opus 5.5.

A comparison puts Gemini 4 Argon at about $1.99 per task versus $3.26 for Astra and $5.98 for Opus, with similar performance at lower cost.

The poster says Argon now ranks first on VALS at 68.9%, ahead of Sonnet 5.5, Opus 5.5, Astra and Sol, while noting benchmarks differ from real-world use.

Argon cut a quantum circuit's qubit count by 40%, freed over 300 TiB of data center memory on its own, is converting an 800K-line C/C++ kernel to Rust, and raises the output limit to 1M tokens.

An investor posts benchmarks of Gemini 4 Argon against GPT-6 Astra and Anthropic Fable 5.1 / Opus 5.5, and speculates Google's internal models often get nerfed after public release.

A chart shows Gemini 4 Argon sitting almost on the Pareto frontier, in the same price-performance range as Claude Opus 5.5 Medium and High.

Google's new flagship Gemini 4 Argon launches first for security defenders only, with a 1M-token output limit, pricing about one-fifth of Astra, and leads 13 of 19 benchmarks.

On Google's own benchmark chart Argon beats Astra, Fable 5.1 and Opus 5.5 in most categories, raises output to 1M tokens and starts at $2/$10 per million tokens, but coding results are mixed and access is limited to the US government and trusted cyber defenders for now.