Anthropic launches Haiku 5.5, says it costs 75% less to run

Anthropic launched Claude Haiku 5.5 on October 7th, saying the small model costs around 75% less to run on average than Haiku 4.5 and calling it the company's cheapest, fastest and most capable small model to date. The announcement on X did not publish per-token prices, benchmark results, a context limit or an API identifier, leaving Haiku 4.5's documented standard rates of $1 per million input tokens and $5 per million output tokens as the only prior reference point. Unconfirmed pre-launch claims from an October 7th X thread by @notjazii put Haiku 5.5 at $0.10 per million input tokens and $0.50 per million output tokens, which would match rather than undercut GPT-6 Luna's listed API rates.

Anthropic launches Haiku 5.5, says it costs 75% less to run The new model is Anthropic's cheapest, fastest and most capable small model to date, according to the company. Its announcement gives no per-token prices or benchmark results. By Ryan Merket https://runtimewire.com/author/ryan-merket · Published · Updated Primary source: X https://x.com/notjazii/status/2107887766524846506 Why it matters The rumored rates would sharply cut Haiku's standard token prices while matching GPT-6 Luna's listed API price. Performance, context limits and launch availability remain unconfirmed, so the thread does not establish a cheaper or better alternative. Anthropic introduced Claude Haiku 5.5 on October 7th, saying it costs around 75% less to run on average than Haiku 4.5. The company called it the cheapest, fastest and most capable small model it has released. The announcement on X https://x.com/claudeai/status/2107894039626277339 does not give per-token prices, benchmark results, a context limit or an API identifier. That cost claim is broader than a token-price comparison. Anthropic's Haiku 4.5 documentation https://platform.claude.com/docs/en/models/haiku-4-5/overview lists standard rates of $1 per million input tokens and $5 per million output tokens. Haiku 5.5's exact rates are still unknown, so the company's average running-cost claim cannot be translated into a new token price from the announcement alone. Before the launch, an October 7th thread by J A Z I I @notjazii https://x.com/notjazii/status/2107887766524846506 predicted a release within about 20 minutes and claimed rates of $0.10 per million input tokens and $0.50 per million output tokens. Those figures remain unconfirmed. They would match GPT-6 Luna's listed API rates https://runtimewire.com/models/native-openai/gpt-6-luna-9a7e60cd1c7bb9d5 , rather than undercut them as the thread suggested. Anthropic had described Haiku 5.5 as a model for high-volume, cost-sensitive applications in its September 28th announcement https://www.anthropic.com/claude-sonnet-5-5 , which said it would join the Claude 5.5 family in the coming weeks. The launch delivers on that timeline. The earlier announcement did not specify Haiku 5.5's price, model identifier, context window or benchmark results. The rumor thread also claimed Haiku 5.5 outperformed Luna and Sol in an unspecified test. Anthropic's launch post publishes no test name, conditions or scores, so that comparison remains unverified. Developers evaluating the new model still need its published rates, limits, availability and evaluation results to judge its cost and performance against alternatives. Haiku 5.5's launch turns the earlier release prediction into news; it does not verify the thread's pricing or performance claims. Anthropic's stated 75% average cost reduction is the only new cost figure in the announcement.