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

A developer questions why Gemini 4 has not been released and asks how pure model companies can compete with firms that already have 4-5B users if its benchmark results hold up.

ValsAI asked Chinese and US models hundreds of politically sensitive questions about China, saying Gemini 4 Argon, Mistral Large 4 and Grok 4.7 occasionally repeat official talking points.

A leak claims Gemini Argon 4 will be released on October 16, just four days after the post.

A new benchmark comparison table lists both Sonnet 5.5 and Gemini 4, and the author reads the rankings through that lens.

A user reports that Gemini 3.8 Flash appears to be routed to Gemini 4 Argon again, and compares Gemini 4 Argon with Muse Wonder on a Minecraft clone task.

The post argues that Argon would have been a decent competitor if Google had released it on announcement day, and that the long delay points to a problem in Google's shipping process.

The author says Meta's new Muse Wonder and Wonder Full models match Claude and GPT on official benchmarks, and complains that Gemini 4 Argon is still unavailable.

The report says Meta enters the frontier model race with Muse Wonder and Wonder Full after Claude Fable, GPT-6 Astra and Gemini 4 Argon, citing benchmark figures for its competitiveness.

A short post calling on Google to release Gemini 4 Argon, reflecting the community's wait for the model.

Artificial Analysis's October 12 leaderboard shows Gemini 4 Argon at 52.6, below Meta Muse Wonder Full's 55.8 and GPT-6 Astra's 52.7.

A video creator says Meta's new model looks benchmark-inflated, with 3D generation weaker than Gemini 3.8 Flash and far behind the Gemini 4 Argon outputs circulating.

The poster rattles off a list of models that can supposedly do the same thing, including GPT-6 Astra Pro, Claude Code Opus 5.5 Max, GLM 5.3, Gemini 4 Argon, Muse Wonder and Grok 4.7.