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

A Chinese post claims Gemini 4 Carbon has launched, leading GPT-6 Astra and Claude Opus 5.5 on nearly all benchmarks and putting Google back on top.

A leak claims Gemini 4 Carbon handles the full workflow from finding vulnerabilities to testing exploits and patching code, with over 2x Gemini 4 Argon's secure-code generation success rate in internal tests.

A leak says Gemini 4 Carbon is priced at $2 per million input tokens and $10 output, matching Argon.

The post says Gemini 4 Carbon scores 59 on the Artificial Analysis intelligence index, beating both Opus 5.5 and GPT-6 Astra for first place.

The poster is surprised Carbon shipped just two weeks after Argon, citing a leading 59 on Artificial Analysis Intelligence Index v5.0 plus gains in long-running tasks and cybersecurity.

Artificial Analysis publishes full benchmark charts for Gemini 4 Carbon running with high reasoning enabled.

A Japanese roundup says Carbon is Argon's successor rather than a higher tier, priced the same at $2/$10 with a future increase, leads the AA Index over Opus 5.5, and is limited to select cyber defenders for now.

A user questions claims that Gemini 4 Carbon beats Opus 5.5 on nearly all benchmarks, suspecting self-run or cherry-picked tests and saying it likely isn't Opus 5.5-level.

Google says Carbon beats another lab's frontier model by over 10x on high-quality security patches, doubles Argon's secure-code success, blocks all prompt-injection attacks, and edges Opus 5.5 59-58 on AA.

A user notes Gemini 4 Carbon arrives just two weeks after Argon and beats it convincingly, seeing Google DeepMind catching up, though it is currently limited to cyber defenders.

A user complains Gemini 4 Carbon, like Argon before it, is withheld from regular users, possibly for another two weeks.

Gemini 4 Carbon scores 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71.0% on CWE-bench, can run for days to root-cause and patch bugs, and ships first to cyber defenders via the Fairwind Program at Argon's price.