Google’s Gemini 4 Pro appears on benchmark platform under disguised name

A model suspected to be Google's Alphabet Inc. ↗ 0.21% next-generation flagship AI, Gemini 4 Pro, has appeared on the Arena benchmarking platform under the name "gemini-3.8-flash," according to developer reports that began circulating on September 17. The model, which carries the internal codename Argon, has not been officially announced by Google, but leaked benchmark results suggest it outperforms both OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1 across coding, reasoning, and agentic tasks.news.aibase+1

What the Benchmarks Show

According to leaked benchmark data shared by developers who tested the model on Arena, Gemini 4 Pro scored 88% on DeepSWE v1.1 for AI-agent coding tasks, 95.3% on Terminal-bench 2.1 for encoding capability, and 86.8% on OSWorld-2.0 for computer proficiency — all reportedly topping scores from Astra and Fable. The model is said to feature a 10-million-token input limit, a 256,000-token output limit, and persistent cross-session memory.news.futunn+1

Developers who tested the model also noted strong performance in SVG and 3D generation. One developer described creating a full creative showcase website in 14 minutes, while others reported the model could generate complete interactive games from a single prompt.news.futunn

On pricing, the model appears positioned as the most cost-effective among the leading frontier models, with reported rates of $2.25 per million input tokens and $11.25 per million output tokens.news.futunn

The Path From Canceled 3.5 Pro to Gemini 4

The appearance of a possible Gemini 4 Pro checkpoint follows a turbulent period for Google's AI model releases. The Wall Street Journal reported in late August that Gemini 3.5 Pro had been scrapped because internal candidates "weren't sufficiently better than the Flash series". Google instead shipped four consecutive Flash models since May, culminating in Gemini 3.8 Flash on September 2.blog+2

Google DeepMind confirmed on July 21 that it had begun "our most ambitious pre-training run yet, for Gemini 4," with CEO Sundar Pichai repeating the announcement on Alphabet's Q2 earnings call the following day. Beyond that, Google has made no official disclosures about Gemini 4 variants, release dates, or specifications.cometapi+1

What Remains Unconfirmed

Google has not acknowledged the Arena testing or confirmed any details about Gemini 4 Pro. Community estimates place a possible public release around October 2026, though no official timeline exists. The distinction between the production Gemini 3.8 Flash — which Google released September 2 with a 1-million-token context window and $0.75 per million input tokens pricing — and the model now appearing under the same name on Arena has fueled speculation that Google is using the label as a disguise for early Gemini 4 Pro evaluation.nokiapoweruser+4

Jasjeet Sekhon, Google DeepMind's chief strategy officer, said at a recent Berkeley event that recursive self-improvement has become "a key component of the rationale behind massive AI investments" — a remark that has taken on new weight amid unverified claims that Gemini 4's pre-training benefited from an RSI closed loop.news.futunn