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

Gemini 4 Argon scores 77.9% vs Opus 5.5's 74.2% on DeepSWE real-code fixes, but trails 57.4% to 66.4% on Terminal-Bench, at half the output price.

A novice investor logs into Google AI Pro via Grok Bot and has it discuss a target with Gemini 4 Argon across rounds to produce a concrete buy plan.

In a Big Technology Podcast interview, Pichai discusses Gemini 4 Argon's ~25 trillion tokens in 72 hours, compute planned 2–3 years ahead, layered safety, and persistent agents spanning Gemini, Search and Gmail.

A user lists the past month's model releases such as Gemini 4 Argon, GPT-6.1 Sol and DeepSeek V4.1, arguing Claude Opus 5.5 still leads.

A post asking readers to pick today's best AI model among Claude Opus 5.5, Kimi K3.1 Pro, GPT-6 Astra and Gemini 4 Argon.

A user spotted a new routing model identifier, gemini-4-argon-d, in Google Antigravity and guesses it may be Gemini 4 Flash.

The poster asks Antigravity users whether Gemini Argon 4 actually solves harder coding tasks or merely does them faster.

A longtime Google skeptic subscribed to Pro and ran Gemini Argon on a Polymarket deep research task, reporting 30+ sources, a math model and rule-by-rule analysis as the best AI output so far.

A short reply claims the first problem can be solved by using Gemini 4 Argon, without specifying what the problem is.

A video describes Jack Wotherspoon's setup where the Interactions API calls Gemini 4 Argon to run multi-agent swarms in remote cloud sandboxes for SEC filings, Python analysis and auto-merging PRs.

The author plans an hourly loop where GPT reviews code, Gemini 4 Argon filters real issues, and Gemini 3.8 Flash fixes them, testing whether Argon can suppress GPT's false positives.

The Google Pro subscription bundled with Pixel grants access to Gemini 4 Argon, Opus 5.5 and Sonnet 5.5, though actual routing varies per account, e.g. Argon maps to Argon Trial and 3.8 Flash splits into Control and Flash-n.