Anonymous gemini-3.8-flash Model Appears on Arena, Benchmarks Lead OpenAI Astra and Anthropic Fable 5.1

An anonymous model named gemini-3.8-flash has recently appeared on the AI model arena Arena. Benchmark tests show the model leads OpenAI Astra and Anthropic Fable 5.1 across multiple capabilities including coding, AI agents, reasoning, and computer operation.

On DeepSWE v1.1, which evaluates AI agent coding ability, the model scored 88.7%, above Astra's 86.9% and Fable 5.1's 69.1%. On the OSWorld-2.0 computer operation test, it scored 86.8%, exceeding Astra's 84.5% and Fable 5.1's 77.9%. On Terminal-bench 2.1, which tests terminal coding ability, the model achieved 95.3%. On GDPval-AA v2, which assesses real-world knowledge work tasks, it scored 2064 Elo.

On pricing, the model is priced at USD 2.25 per million input tokens and USD 11.25 per million output tokens. Astra is priced at USD 12 per million input tokens and USD 60 per million output tokens, while Fable 5.1 is priced at USD 10 and USD 50 per million tokens, respectively.

Google's flagship AI model Gemini 3.5 Pro was originally scheduled for launch in June 2026, but the release plan has been delayed by several months, and its website still displays "3.5 Pro coming soon." Reports say Google has cancelled the launch of Gemini 3.5 Pro because the model's improvement was limited and its performance was even weaker than the Flash version. Google last launched Gemini 3.1 Pro on February 19, 2026, seven months ago. In early September 2026, Google released "Gemini 3.8 Flash," and the anonymous model appearing on the Arena platform shares the same name.

Jasjeet Sekhon, chief strategy officer at Google DeepMind, said at an event in Berkeley that recursive self-improvement has become a key part of the logic behind massive AI investment. On September 14, 2026, a Google study named Dream-RSI was made public, examining how to enable agents to continuously improve their own search strategies during exploration and to continue upgrading the next round of exploration from experience.