◆ THE LAB · Models
Anthropic's cheaper Haiku 5.5 turns text into more tokens
The News
Anthropic released Claude Haiku 5.5 on October 7, 2026, and it is available now. For prompts up to 100,000 tokens, input costs $0.10 per million tokens and output $0.50, a tenth of Haiku 4.5's $1 and $5. Allowing for pricier long prompts and a new tokenizer, Anthropic says it costs around 75% less to run on average.
Chandni Melwani
Founder & Editor
Oct 8, 2026 · 1 MIN READ
Know More
- — Context window: 1M tokens, with up to 128k output tokens, up from 200k and 64k on Haiku 4.5.
- — Prompts over 100,000 tokens: $0.50 input and $2.50 output per million tokens, half Haiku 4.5's rates.
- — Breaking changes include: non-default sampling parameters, assistant-message prefill and manual extended thinking now return errors.
- — Availability: now on all platforms, Anthropic says, including Amazon Web Services, Google Cloud and Microsoft Azure.
Anthropic explains the 75% in footnote 2 of its October 7 launch page. Haiku 5.5 is priced 90% below Haiku 4.5 for requests up to 100,000 tokens and 50% below for anything larger, and on Haiku 4.5, 90% of requests fell into the first group. Anthropic’s calculation also accounts for how many tokens a given piece of work uses. Haiku 5.5 has an updated tokenizer, which Anthropic says “means it uses slightly more tokens per task.”
The developer docs measure input text, and say the same text produces approximately 30% more tokens on Haiku 5.5 than on Haiku 4.5, with the exact increase depending on the content. Anthropic lists the change with an action attached: “Recount prompts and revisit max_tokens and cost estimates.”
Sonnet 5.5 and Opus 5.5 remain better choices for complex agentic coding tasks, Anthropic says, and its table puts Haiku 5.5 at 39.2% on Terminal-Bench 4.0, against 70.6% for Sonnet 5.5.
The Read
Anthropic says Haiku 5.5 is best suited to more narrowly scoped tasks that might have cost too much on earlier Claude models, and that it pairs well with Opus 5.5 and Sonnet 5.5 as a subagent on coding work. Devin CLI users can pair it with Opus 5.5. That points to more coding agents pairing a lead model with Haiku.
Room for Disagreement
Four of the six customers Anthropic quotes don't mention a lead model. AlphaSense compared Haiku 5.5 with Haiku 4.5 on a feature making about 8M calls a week, and Asana with the model it uses today. Those quotes point to Haiku 5.5 replacing the model a product already runs.
Related
Frequently Asked Questions
Which rival is in Anthropic's launch benchmark table?
OpenAI's GPT-6 Luna, the only model from another company in it. In Anthropic's own table, Haiku 5.5 has a partial-credit score of 72.4% to GPT-6 Luna's 48.9% on the offline subset of OSWorld 2.1, a test of operating a real computer, and an Elo rating of 1,620 to 1,437 on GDPval-AA v2.1, Artificial Analysis's test of real-world professional work across 44 occupations. Anthropic ran GPT-6 Luna on OSWorld itself, on the same 82 tasks through OpenAI's API. Haiku 5.5's scores are at max effort unless Anthropic notes otherwise.
What does Haiku 5.5's effort setting do?
It lets users decide whether to optimise for cost or intelligence, and Anthropic says Haiku 5.5 is its first Haiku-class model to have one. Adaptive thinking, where the model decides when and how much to think, is on by default. Thinking tokens count toward max_tokens, so a small limit can stop after a thinking block and before any text.
What else did Anthropic change on October 7?
It halved the price of cache reads on Claude Sonnet 5.5, to $0.10 per million tokens from $0.20, which it says cuts Sonnet 5.5's cost on most agentic tasks by around 20%. It also announced a monthly API credit for use on the Claude Platform, which it said it would roll out to all Max and Team subscribers in the week of October 7: $100 for Max 5x users, $200 for Max 20x users and up to $500 for Team subscribers, pooled across their users.
Sources
- Anthropic, 'Introducing Claude Haiku 5.5' (October 7, 2026): the launch date, the pricing table for Haiku 5.5, Haiku 4.5 and Sonnet 5.5, footnote 2 behind the 'around 75% less to run' average, the benchmark table, the effort setting, the customer quotes, the Sonnet 5.5 cache-read price and the Max and Team API credit
- Claude Platform Docs, 'What's new in Claude Haiku 5.5': the 1M context window and 128k output limit, the tokenizer change (approximately 30% more tokens for the same input text), adaptive thinking and effort, and the breaking changes
- Anthropic, 'System Card: Claude Haiku 5.5' (October 7, 2026), section 8: the capability table, the max-effort configuration for Haiku 5.5's scores and how the GPT-6 Luna figures were obtained
Chandni Melwani
Chandni Melwani is the founder and editor of New in AI, covering AI agents, M&A, and enterprise adoption. She holds a Master's in Management of Artificial Intelligence from Queen's University and brings a practitioner's perspective from her work in Data and AI leadership.
Follow ↗The Lab Briefing
Get the five-beat briefing in your inbox.
The big story, frontier labs, AI at work, e-commerce and the capital crossing borders, each with the part that tells you why it matters, not just that it happened.
Free · No spam · Unsubscribe in one click
More from The Lab
Labs
Anthropic is turning away distillers' paying traffic
Anthropic says early evidence suggests its new defences have cut the data harvested by distillers, who covertly extract Claude's capabilities to replicate them in other models, by around 90 to 99%, and that it has turned away meaningful revenue and trillions of tokens' worth of paying traffic from distillers.
Chandni Melwani · Global · Oct 21, 2026
Models
Meta releases Muse Spark Open under its own licence
Meta's open-weights Muse Spark Open is free of charge under its Muse Model License, but if a company or any of its affiliates offers AI models to third parties via API or a similar interface, and together they gross over $20m in any consecutive 12 months, the company must notify Meta and sign a separate written agreement before offering the model or a derivative as part of that model-as-a-service business. Meta's own table puts it behind GLM 5.3, Kimi K3 and Qwen3.8-2.4T-A95B on the Artificial Analysis Intelligence Index.
Chandni Melwani · Global · Oct 20, 2026
Models
Google's Argon beats Opus 5.5 on LiveBench reasoning, not coding
Gemini 4 Argon costs $2 per million input tokens and $10 per million output through November 30, then $4 and $20, the same as Claude Opus 5.5. On LiveBench 2.0 it beats Opus 5.5 on reasoning but trails it on coding, and it ranks sixth on Vals' Terminal-Bench 4.0.
Chandni Melwani · Global · Oct 18, 2026