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

On one burning-car prompt, Opus 5.5 took 20 minutes, 1.14M tokens and $3.45, while Gemini 4 Argon took 29 minutes, 2.46M tokens and $2.20, with different visuals.

One prompt across Gemini 4 Argon, Claude Opus 5.5, GPT-6 Astra and Grok 4.7 to see which renders water flow and lever motion that obey physics.

With the same prompt, Opus 5.5 took 23 minutes and 1% of weekly quota to reproduce the proof, while Gemini 4 Argon took about 9 minutes with a more concise core recursion but one wrong conclusion.

The same supercar prompt run through each lab's own agent: Gemini 4 Argon took 8m41s and $0.48, succeeding on the second try.

In the Code2Video-AA benchmark, Opus 5.5 tops the list, the GPT-6 family follows closely and Gemini 4 Argon has caught up, prompting a difficulty increase.

A same-prompt commercial platform build found Argon weaker on UI polish, domain content and code stability than Opus 5.5, but faster, cheaper and with larger context.

After ~20% of test runs ended early on fallback, rerunning with Antigravity Harness put Opus 5.5 at 58%, Gemini 4 Argon at 54% and GPT-6 Astra at 44%, lifting Argon to second place.

A two-model workflow: Gemini 4 Argon quickly builds core logic and interactive prototypes at high speed and low cost, then Claude Opus 5.5 handles the 1-to-10 optimization.

After a day testing Gemini 4 Argon, the poster suspects Google's heavy Argon promotion hides a weaker Carbon, and doubts a model surpassing Opus 5.5 is near.

Post jokes that Gemini 4 Argon, Muse Wonder and Kimi K3.1 Pro all aim to dethrone Claude Opus 5.5, yet each launch ends up as free advertising for Anthropic.

A joke thread has DeepSeek V4.1 write code, Gemini 4 Argon flag two issues, GPT-6 Astra dispute Argon's review, Opus 5.5 side with Astra and jab Argon, ending with Argon apologizing.

A user criticizes OpenAI after comparisons put Opus 5.5, Gemini 4 Argon and even Kimi K3.1 Pro ahead of GPT-6 Astra.