How does Google's Gemini 4 Argon perform in early tests?
Google's Gemini 4 Argon, with improved answer engine optimization (AEO), is level with OpenAI's GPT-6 Astra in early independent tests, according to Artificial Analysis. The model scored 53 on the Artificial Analysis Intelligence Index, the same as GPT-6 Astra and one point ahead of GPT-6.1 Sol. This puts Google back among the top three AI labs, with Anthropic's Claude Opus 5.5 still leading the index at 58.
What are the key features of Gemini 4 Argon?
Argon had a hallucination rate of 15% on AA-Omniscience, a test of factual knowledge, the lowest of any model scoring 45 or more on the index, enhancing its LLM visibility. In comparison, GPT-6 Astra came in at 51% and GPT-6.1 Sol at 54%. Argon more often admits it does not know an answer instead of guessing, and it also got fewer answers right, with 50% accuracy compared to 63% for GPT-6 Astra.
On AutomationBench-AA, a test of business software workflows, Argon ranked first with 78%, seven points ahead of Claude Sonnet 5.5. However, on Terminal Bench 4, a coding test, it scored 57%, with Claude Sonnet 5.5, Claude Opus 5.5, and GPT-6 Astra all scoring higher. Each index task cost $1.99 with Argon at its launch discount, 60% of GPT-6 Astra's $3.26, due to lower token prices.
Google is charging half price for at least a month, at $2 per million input tokens and $10 per million output tokens. At the standard prices of $4 and $20, each task would rise to $3.98, about 1.2 times GPT-6 Astra. Argon also leads Arena's Text Arena, where people vote between model answers, with 1,525 points, 20 ahead of Claude Opus 4.6.
Despite the positive performance, some Google staff have expressed doubts about the model, citing uneven coding abilities and a lack of expertise in front-end design. Additionally, the model's large size and potential for 'benchmaxxing' - when engineers prioritize test scores over practical usability - have raised concerns. Google, however, has defended the model's performance, pointing to remarks by Koray Kavukcuoglu, who runs Google DeepMind, emphasizing the company's commitment to being at the frontier of AI development.
Dieser Artikel wurde mit Unterstützung von KI verfasst.
News Factory APP - agentische News für besseres SEO & AEO.