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

A leak claims Gemini 4 Flash is in internal testing under the codename Hydrogen, and could pressure OpenAI if it nears Argon at a lower price.

A Japanese user compares Opus 5.5, GPT-6 Astra and Gemini 3.1 Pro on one task, finds Gemini worst, and buys the Gemini 4 Argon add-on.

A post claims a gemini-3.8-flash-router appeared in Antigravity ACP two days ago and was speculated to be Gemini 4 Flash, now reportedly named Gemini 4 Hydrogen.

A Chinese user asks what happens to the US model race if Gemini 4 Flash, codenamed Hydrogen, really beats GPT 6.1 Sol.

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.

A leak says the Gemini 4 Flash-tier model is codenamed Hydrogen, may launch this week, and is seen as a rival to GPT-6.1 Terra.

A post traces Google's element-based naming—Argon internally codenamed Barium, Carbon in testing, a Flash model possibly called Hydrogen—and guesses Helium comes next.

A leak says Gemini 4 Flash is in internal testing as "Hydrogen", a gemini-3.8-flash-router model ID appears in Antigravity, and launch may come this week or next.

A user predicts Gemini 4 Flash, internally codenamed Hydrogen, will outperform GPT-6.1 Sol, with no data offered.

A post rounds up rumored Gemini 4 codenames—Carbon said to target Opus 5.5 and Hydrogen close to Argon but far cheaper—and asks where that leaves OpenAI and DeepSeek V4.1 Flash.

A weekly roundup says Gemini 4 Argon rolls out to Antigravity IDE users within 24 hours, alongside AA Intelligence Index v5 and Codex Terra news.

LuminaBench says a new Gemini 4 Flash tier reportedly codenamed Hydrogen could, if close to Argon but much cheaper, let Google take second place from OpenAI temporarily.