Claude Haiku 5.5
Blog · Watch/Read · (gathered 2026-10-14)
- Simon Willison evaluates Anthropic’s new Claude Haiku 5.5, positioning it as a direct competitor to OpenAI’s GPT-6 Luna, particularly focusing on its pricing structure and performance.
- Haiku 5.5 matches Luna’s price ($0.10/$0.50 per million tokens) for prompts under 100,000 tokens, but becomes significantly more expensive ($0.50/$2.50) beyond that, unlike Luna’s more gradual increase. A “less generous tokenizer” also means a hidden 1.25x price increase compared to Haiku 4.5.
- While Haiku 5.5 shows improved benchmark scores and drawing capabilities (e.g., the “pelican riding a bicycle” test) compared to its predecessor, it defaults to “medium” reasoning and doesn’t allow disabling it, impacting cost and generation time for complex tasks.
- Anthropic is offering “really generous” monthly API credits ($100-$500) to Max and Team subscribers, matching their subscription cost, which can be managed with a new auto-reload disable feature to prevent billing surprises, though credits do not roll over.