Leaked Gemini 4 Pro benchmarks beat GPT-6 Astra and Claude

A model developers believe is Google's unannounced Gemini 4 Pro appeared on the Arena benchmarking platform on Sept. 17 under the name "gemini-3.8-flash," reportedly codenamed Argon. Leaked scores of 88% on DeepSWE v1.1, 95.3% on Terminal-bench 2.1 and 86.8% on OSWorld-2.0 put it ahead of OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1. Google has not confirmed the testing.

TechBriefly Open source

Leaked Gemini 4 Pro benchmarks beat GPT-6 Astra and Claude

Reported 10M-token input and 256K-token output limits with cross-session memory
Leaked pricing of $2.25 per million input tokens and $11.25 per million output tokens
Testers say it built a showcase website in 14 minutes and full games from one prompt
Google has not disclosed Gemini 4 variants, release dates or specs

#Google#Gemini#OpenAI#Anthropic

Leaked Gemini 4 Pro benchmarks beat GPT-6 Astra and Claude

Fastly launches AI Runtime Control and AI Firewall for enterprises

Leak: Gemini 4 Pro Enters Arena Testing for Multimodal AI

10 days that changed the course of AI: labs call for slowdown

Google tests Gemini-powered Call for me in Phone app