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

A post says Google's shared Gemini 4 Argon benchmarks greatly outperform Opus 5.5 and GPT-6 Astra, and mentions a new internal checkpoint named Carbon.

The author shares a comparison chart and says the new Argon checkpoint, possibly named Carbon, is a significant improvement over the earlier Barium-B version.

A user expresses surprise that Google is already testing the next-generation Gemini 4 Carbon before Gemini 4 Argon has even been released.

The poster claims Gemini 4 Argon was quietly routed from 3.8 Flash about a month ago, and that Google has built hype by not releasing what is now the most anticipated unreleased model.

A Japanese post notes that timelines are full of claims that Gemini 4 Argon is actually routed to Gemini 3.8 Flash, which would be a massive disappointment if true, and jokes that its compliments are half-hearted.

A sarcastic post says the odds of Gemini 4 Argon getting a public release next week are lower than the odds of achieving AGI.

A video claims gemini-3.8-flash-exp-a in Antigravity appears to route to Gemini 4 Argon, and that Gemini 3.1 Pro now also points to 3.8 Flash, an unconfirmed finding.

The poster suspects Gemini 4 Argon will be passed off as a rebranded Gemini 4 Flash, repeating an earlier pattern of premature celebration.

A post says Gemini 4 Argon leads the Vals Index and DeepSWE benchmarks, but its full capabilities are only available to cyber defenders.

Google employees are reportedly testing a new Gemini 4 version called Carbon internally, rating its coding ability close to Opus 5.5, while Argon has not yet been released publicly.

A user posted a "pelican riding a bicycle" generation test for Gemini 4 Argon, reportedly Gemini 4 Pro, with the link in the comments.

A developer says he is looking forward to Qwen4-27B more than Gemini Argon, suggesting limited enthusiasm for Google's new model.