Anthropic has shipped Claude Haiku 5.5, its first Haiku-tier update in about a year, positioning it as a low-cost workhorse for high-volume tasks like summarization, compactions, and database queries. According to Latent Space's AINews roundup, Anthropic calls it "the cheapest, fastest, and most capable small model we've ever released," and it is priced to match OpenAI's GPT-6 Luna. The same day, Anthropic cut Sonnet 5.5 prices and added API credits to subscription plans.
Independent evaluation from Artificial Analysis gives Haiku 5.5 an Intelligence Index of 43 at max effort, ahead of GPT-6 Luna (38), Gemini 3.8 Flash (41), and GLM-5.3 Flash (42), while trailing Claude Sonnet 5.5 (56). The main caveat is token usage: at max effort Haiku 5.5 consumes about 162K output tokens per Index task, roughly three times Luna's ~50K. At equal scores, Haiku is still more verbose, and cost-per-task figures are provisional because the 5x price step above 100K tokens is not yet modeled.
The source notes some discrepancies between vendor claims and independent results. Anthropic reports TerminalBench at 39%, while Artificial Analysis measures 33% on Terminal-Bench 4.0. A pre-release safety bug causes the model to over-refuse, which Anthropic says it is fixing; Artificial Analysis expects its AutomationBench score to rise after a re-run. Haiku 5.5 is the first Haiku with Anthropic's effort settings and adaptive thinking, and it expands context to 1M tokens, up from 200K in Haiku 4.5.