Anthropic releases Claude Haiku 5.5 at a tenth of Haiku 4.5's per-token price

Anthropic's Claude Haiku 5.5 costs $0.10/$0.50 per million tokens under 100K, plus a Sonnet 5.5 cache-read cut and monthly API credits for Max and Team.

Anthropic has released Claude Haiku 5.5 (October 7, 2026), a small model it pitches for high-volume work: summaries, compaction, classification and subagent tasks under Opus 5.5 or Sonnet 5.5. If you run cheap calls at scale, the price math changes. For prompts up to 100K tokens it costs $0.10 input and $0.50 output per million tokens, against $1 and $5 for Haiku 4.5. It is also the first Haiku with an adjustable effort setting.

The "75% cheaper" figure is Anthropic's blended average, not the per-token cut. Per-token rates are 90% lower below 100K tokens and 50% lower above it, and the new tokenizer counts more tokens per task. ComputingForGeeks measured 1.11x to 1.36x more tokens on the same files. It also notes that the 100K boundary is counted in Haiku 5.5 tokens, so an 80K prompt on Haiku 4.5 can land in the pricier tier. Temperature, top_p and top_k now return a 400 error.

Treat the benchmarks as vendor-reported for now. Anthropic lists 39.2% on Terminal-Bench 4.0. Latent.Space reports that Artificial Analysis measured 33%, and that the model uses more tokens than GPT-6 Luna at equal scores. Simon Willison points out that above 100K tokens, Luna's pricing looks better.

Two other changes shipped the same day. Sonnet 5.5 cache reads are halved to $0.10 per million tokens, which Anthropic says cuts most agentic runs by about 20%. Anthropic also says it is rolling out monthly API credits this week: $100 for Max 5x, $200 for Max 20x and up to $500 pooled for Team. The credits docs say they cover the Claude API and Agent SDK but not Claude Code or the cloud marketplaces.

What to do: re-count your prompts with the new tokenizer and benchmark your own subagent step before switching.

Sources