Google launched Gemini 4 Argon on Wednesday as its most capable model for coding, office tasks, and cyber defense. The rollout arrived one day after GPT 6.1 Sol and one week after Claude Opus 5.5.
On the DeepSWE v1.1 software engineering test, Argon achieved a score of 77.9%. That result outperformed Claude Opus 5.5 at 74.2%, GPT-6 Astra at 74.1%, and Claude Fable 5.1 at 67.4%. Earlier in July, Gemini 3.6 Flash recorded a 49% score on the same benchmark.
Google also expanded the model's output capability to 1 million tokens in a single response, up from 64,000 tokens. Based on an estimate of three-quarters of a word per token, this increases maximum response length from about 48,000 words to roughly 750,000 words.
For traders holding positions on Bitget, Bybit, MEXC, or OKX, these AI benchmark results change nothing directly regarding funding rates, holding costs, or exchange liquidity. The developments remain isolated to tech benchmark rankings rather than crypto derivative pricing.
Source: decrypt — Gemini 4 Is Here, and Google’s Flagship Tops All Other AI Models on Cybersecurity
Price today: Solana