LLM benchmarks

Google has released Gemini 4 Argon, its first frontier model in over seven months, featuring a 1 million output token limit — an industry first. Independent benchmarks show it matching OpenAI's GPT-6 Astra but trailing Anthropic's Claude Opus 5.5.

OpenAI has begun rolling out GPT-6 Astra, priced at $10/million input and $50/million output tokens to match Claude Fable. The model claims a 99.9% score on ARC-AGI 3 using a custom harness and leads on security and long-context benchmarks, though it trails Fable on general intelligence rankings.

Ai2 has released BenchMIRT, a technique that uses multidimensional item response theory to analyze which underlying capabilities drive scores on individual benchmark questions. Trained on 100 LLMs across 16 benchmarks and 34,000+ questions, it found that benchmarks like BBQ and WMDP measure general reasoning more than safety, despite being marketed as safety evaluations.