Claude's new model released! Benchmark scores crush GPT-6 Luna, priced even cheaper than Liang Wengu, OpenAI can only give away reset cards to save face.

What happened: Anthropic released Claude Haiku 5.5, a small model that beats DeepSeek V4.1 Flash and GLM-5.3-Flash on benchmarks and wins item by item against GPT-6 Luna, at matching price. Key numbers: Haiku 4.5 launched a year ago at 10x the price of 5.5. Requests under 100k tokens get 90% off input and output; above that, 50% off. Why it matters: Anthropic positions Haiku 5.5 for execution-layer tasks, not as a Sonnet 5.5 replacement, and recommends Sonnet 5.5 or Opus 5.5 for complex agent coding. Migration requires five breaking API changes.

Claude's new model beats GPT-6 Luna on benchmarks at lower price.

Claude's new model beats GPT-6 Luna on benchmarks at lower price. — impact context not separately stored; see What changed.