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

A teaser post says three new models will face off, naming Gemini 4 Argon as launching alongside GPT-6.1 Astra and Claude Fable 5.5.

A leak claims Antigravity will add Opus 5.5 and Sonnet 5.5 with Low, Medium and High reasoning effort, that GPT-OSS 120B may be phased out, and that Gemini 4 Argon could fill the slot.

A leak says Google dropped Gemini 3.5 Pro and moved straight to Gemini 4 Argon, now open to select partners and positioned against Opus 5.5 and GPT-6 Astra.

A post says the user is praying Gemini 4 Argon is better than Sonnet while being almost unlimited even on the $100 tier.

A video post claims Gemini 4 Argon outscores an Anthropic model that has already been replaced twice, joking that AGI is confirmed, though it offers little actual information.

A post invites readers to vote on the best model for coding among GPT-6 Astra, Gemini 4 Argon and Claude Opus 5.5.

A Japanese weekly AI roundup covers GPT-6.1 Astra being held back over safety, Gemini 4 Argon supporting 1M-token output, and the free Gemini tier dropping to Flash-Lite only from 10/9.

A poster notes that Google's Antigravity platform just added Opus 5.5 and Sonnet 5.5, and speculates that Gemini 4 Argon could be the next model added.

Given the same prompt, the four models produced mechanical bee videos: Opus 5.5 had the clearest detail, Sonnet 5.5 was the most surprising, Astra was unremarkable and Gemini 4 took a different but acceptable direction.

The post says Gemini 4 Argon launched on September 30 with top benchmark scores and $2 per million input tokens, but Bloomberg reported that some inside Google find it less impressive in real work, especially coding, and the stock gave back most of its gains.

An Arabic-language AI weekly roundup names Gemini 4 Argon alongside GPT-6.1 Sol and Claude Sonnet 5.5 as the week's major large models.

Commenting on Google AI subscription changes, the user argues free and Plus tiers should not include Pro models, and that a $5 plan would only cover about two prompts even with Gemini 4 Argon access.