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

A post reports recurring model-unavailable errors and speculates Gemini 4 Flash is coming at roughly half Argon's price, possibly beating Sonnet 5.5.

In the latest AA Intelligence Index, GPT-6.1 Terra leads with 51, Gemini 4 Argon is second at 50 and Claude Fable 5.1 third at 49, though the author says actual outputs look off.

A video notes Terra 6.1 (51) edges Argon (50) and Fable 5.1 (49) on the AA Intelligence Index, but says real-world performance differs.

After a notice that chats may be human-reviewed with opting out requiring deleting chats, the author downgraded their Gemini account and immediately lost Argon access, criticizing Google's treatment of users.

A one-shot GPT-6.1 Terra demo reportedly scores one point above Gemini 4 Argon on the new Analysis Intelligence Index v5.

The author used Gemini 4 Argon to build a preview animation where anime versions of Dario and Claude explain a drug development pipeline.

A Microsoft AI diagnostic study published in Science is cited as another independent validation of Gemini 4 Argon's promising health capabilities.

Bindu Reddy reads the new LiveBench results as Fable 5.1 and Astra leading, Argon beating Terra on data analysis, and Google topping open source for the first time.

A multi-model review uses Gemini 4 Argon as a reference point, noting Kimi K3.1 Pro edges past Argon's ~53 on Artificial Analysis.

A tutorial claims you can set up a cheap compliant Gemini subscription in under 10 minutes to use Gemini 4 Argon and Claude Opus 5.5 in Antigravity for about 30 RMB a month, renewable for four years.

AkitaOnRails says their benchmark agrees with another result; their LLM benchmark v4 compares Gemini 4 Argon, GPT-6.1 Sol, Kimi K3.1 Pro and Claude Sonnet 5.5.

In the overhauled AI Intelligence Index, Claude Opus 5.5 leads at 58 while Gemini 4 Argon scores 50 in fourth, behind GPT-6 Astra and GPT-6.1 Sol; 60% of the index uses private or held-out evals.