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

A tip post notes Google AI Studio's free API key gives no-cost access to Gemini 3.8 Flash, while flagship Gemini 4 Argon is excluded from the free tier and free-tier data is used to improve products.

The updated index adds FrontierCode 2.0 and Terminal-Bench-Science; Claude Opus 5.5 leads at 58, ahead of GPT-6 Astra (57) and GPT-6.1 Sol (55), with Gemini 4 Argon (high) next at 50.

After Artificial Analysis rebuilt its Intelligence Index, Opus 5.5 scores 58, GPT-6.1 Terra 51 and Gemini 4 Argon 50, which the poster finds odd.

A Japanese user shares a full day of hands-on time with Gemini 4 Argon, with an image delivering a wry one-word conclusion.

Gemini 4 Argon is now fully rolled out with ample quota in its own separate pool, plus a tip for switching to Antigravity inside the Gemini client.

A user had Gemini 4 Argon reverse-engineer a prompt from a screenshot, then fed it to Banana 2.1 and got a near-identical image, raising doubts about how exact the match can be.

A user accepts Opus 5.5 at #1 but is puzzled that GPT-6.1 Terra ranks above both Claude Fable 5.1 and Gemini 4 Argon.

After Artificial Analysis upgraded its Intelligence Index to v5.0, several models dropped sharply while Gemini 4 Argon's score rose under the new standard and GPT-6 Astra also climbed.

A user posts a video of Gemini 4 Argon completing a task and calls it bad news for GPT-6.1 Terra.

The user floats a hot take that DeepSeek is better than Gemini 4 Argon and asks for agreement, without citing specific benchmark data.

A researcher says they feed drafts and paper excerpts to Gemini Argon with a simple 'help me organize this' prompt.

A Google developer account says Gemini 4 Argon in Antigravity handles system-level tasks, citing 2x speedup on legacy research code and 40% less session prep overhead.