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

Gemini 4 Argon scores 55.0% on FrontierSWE, close to Fable 5.1 and behind only Opus 5.5 and GPT-6 Astra, with large gains over Gemini 3.8 Flash.

A poster says Gemini 4 arrived hours after they said Google needed a frontier model, and questions whether its 1M output tokens and benchmark numbers will hold up in real use.

On Artificial Analysis, Opus 5.5 leads while Gemini 4 Argon matches GPT-6 Astra's score at $1.99 per benchmark task, 39% cheaper than Astra and 67% cheaper than Opus.

A post says Gemini 4 Argon outscores GPT-6 Astra and Claude Opus 5.5 on most benchmarks, leads legal tasks by 19.6%, raises output to 1M tokens, and is limited to Google staff, vetted security groups and trusted testers.

A Japanese post says Gemini 4, codenamed Argon, beats Astra and Fable on benchmarks, though general availability is still some way off.

French outlet Frandroid says Google has announced Gemini 4 Pro, its next high-end model, following OpenAI and Claude; no release date is given, but early benchmarks look promising.

A post lists Argon agent work such as freeing over 300 TiB of data-center memory, beating a quantum algorithm baseline by 40% and porting the Fuchsia kernel to Rust, with an intro price of $2/$10 per 1M tokens that doubles later and no public access yet.

A Spanish-language account says Google's Gemini 4 Argon, described as its most powerful model, matches GPT-Astra and Claude Opus in coding and cybersecurity and will launch in phases, first for enterprises.

Gemini 4 Argon scores 53 on the Artificial Analysis Intelligence Index, tied with GPT-6 Astra and Fable 5.1 and one point ahead of GPT-6.1 Sol.

Sources tell reporters there is a gap between Gemini 4's benchmark scores and its performance on real-world tasks, including some coding work, and internal opinions are mixed on whether it has caught up to OpenAI and Anthropic.

A Chinese post rounds up Argon's results: an AA index of 53 matching GPT-6 Astra, a new DeepSWE SOTA at 77.9%, a 1M-token output limit and a per-task cost at 60% of Astra's.

The post clarifies that Argon is only a rename of Gemini 4 Pro, currently limited to a small group of testers, and says Google prepared well and will open it widely soon.