Articles tagged AI benchmark

Claude Sonnet 5.5 lands two points from Opus 5.5 on real-work benchmarks and cuts the cost of a task by up to 30 percent.

Across six Vending-Bench runs the Astra agent closed at 15,515 dollars against 5,422 for Claude Fable 5.1, and cleared all five Drone-Bench tasks.

DeepSeek V4.1-Flash drops to 890 bytes of cache per token under an MIT licence, with flat API pricing and a one million token context.

Artificial Analysis rewrote its Intelligence Index after GPT-6 Astra scored 61, adding two fresh benchmarks and lifting private data to 40 percent.

Our GPT-6 Astra test: the record score comes from a custom harness, and caching decides the invoice far more than the ranking does.

Our Claude Fable 5.1 test: headline rate unchanged, cache reads quartered, and three API breaks to plan for before migrating.

GPT-6 Astra ships at 10 dollars per million input tokens, with a 99.9% ARC-AGI-3 score that drops to 62.7% on the standard harness.

Our V4-Flash-Vision test: three wins out of eleven against Opus 4.8, six text benchmarks up, and twelve points behind on repository-scale code.

Ox Alpha ships free with a million-token context window from an anonymous provider that retains your prompts. What it is worth and what it costs.

Grok 4.6 ties GPT-5.6 Sol Max on the intelligence index and bills two dollars per million tokens, more than 60 % under Claude Opus 5.