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

A leaker says Theo is teasing a Fable 5.5 release today, and the poster plans to compare Opus 5.5, Fable 5.5, Astra 6 and Argon.

A Japanese user warns against mistakenly subscribing to the ¥725 AI Plus tier, which lacks both Gemini 4 Argon and the Tasks feature, arguing Pro works out cheaper.

The author fed a leaked 3D video from the AA arena into Gemini 4 Argon to reverse-engineer the prompt and recreate it, saying the result came close to the original after two rounds of tweaks.

A user posts a video saying it was generated with Gemini 4 Argon, showing the model's video and multimodal output.

A Japanese user ran the same 20 prompts through Gemini 4 Argon, 3.8 thinking mode and 3.6 Flash, finding clear gaps on creative tasks while only 3.8 thinking mode made factual errors.

A post with an image asks whether Gemini 4 Argon is the new king after its successful launch, and jokes about whether Sundar Pichai actually pulled it off.

A post outlines a four-command REA plus Gemini 4 Argon workflow it says can take a company from recon to clean-room rebuild, and notes most people still can't access it.

A post collects the most cost-effective ways to use Gemini 4 Argon plus tutorials for getting the most value, noting Colab compute units are paid-only and not included in Google's recent six-month offer.

A new benchmark has models run a Stronghold Crusader castle on their own, with Gemini 4 Argon scoring 62.3 ahead of GPT-6.1 Sol and Claude Opus 5.5, against a human score of 100.

A Japanese first impression of Gemini 4 Argon describes it as finishing its reasoning and delivering nothing, joking that it is like a sage who only meditates.

A post says Gemini 4 Argon can be called at zero cost on unorouter.com by signing up for a key, though it is often at capacity and needs retries.

A user posted an image saying Gemini 4 Argon's performance genuinely deserves its current ranking position.