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

A video compares Gemini 4 Argon in Antigravity against GPT-6 Astra in Codex, claiming Argon wins 54 to 44 on FrontierCS 2 and calling Astra tasteless.

A comparison of $20 AI coding subscriptions finds Google AI Pro bundles Gemini 4 Argon and Flash, against Codex Plus, Claude Pro and Cursor Premium.

A user who spent a day driving Gemini 4 Argon in Antigravity says it needs excessive iteration and guidance, and credits Claude Code desktop and Opus 5.5 for his productivity.

Induction Labs adds Gemini 4 Argon, DeepSeek V4.1 Flash and MiniMax M3.1 Flash to its leaderboard; Argon matches GPT-6.1 Sol but sits behind the frontier.

One prompt asks Claude Opus 5.5, GPT-6 Astra, Gemini 4 Argon and Grok 4.7 to animate a working gear pump; the author posts unedited outputs and lets viewers judge which actually moves fluid.

Across 20 expert-written report tasks, scores ranged 54.3%–64.1% with Gemini 4 Argon highest; most lost points came from deciding what to include and explaining what facts mean.

A Google AI Pro user finds Gemini 4 Argon's Medium thinking level sufficient for most tasks, with High prone to stalling, and suggests Google add an auto mode that picks thinking intensity.

A screen-recorded reaction in which the author uses Google Argon, reportedly Gemini 4 Pro, to walk through Anthropic's full S-1 registration statement.

A reply shows a user asking Gemini to update a graph labeled with Argon's high setting, touching on Gemini 4 Argon tier info.

Google says Gemini 4 Argon lifted gemma-4-E2B-it from ~37,900 to 108,000+ tokens/sec on an A100 via vLLM with 100% token parity and no quality drop, across 8 PRs.

After two days of trying Gemini Argon as a workflow advisor and failing, the author argues it lacks big-model knowledge density and is likely a small Flash-tier model, with larger releases to come.

A weekly roundup says Kimi K3.1 Pro matches Claude Opus 5.5 and GPT-6 Astra on several benchmarks and on Arena agentic tasks trails only Gemini 4 Argon and Muse Wonder Full.