Claude Haiku 5.5 is out. Anthropic released it on October 7, 2026, and the @claudeai launch thread is timestamped 18:01 UTC. Anthropic calls it “the cheapest, fastest, and most capable small model we’ve ever released”. The model ID is claude-haiku-5-5, and its model page reads “Released October 7, 2026”.
This page opened on September 23 as a tracker for a model Anthropic had named but not dated. It now holds the release record and grades every claim that came before it. The headline is the price: $0.10 per million input tokens on prompts up to 100,000 tokens, against $1 for Haiku 4.5.
On this page · 12 sectionsOpen
Claude Haiku 5.5 shipped on October 7, 2026, posted by Anthropic at 18:01 UTC, 15 days after Anthropic said it would follow Opus 5.5 in the coming weeks. The model ID is claude-haiku-5-5, with a 1M token context window and 128K max output.
It lists at $0.10 per million input tokens and $0.50 output for prompts up to 100,000 tokens, and $0.50 and $2.50 above that. Haiku 4.5 is $1 and $5. Anthropic says it costs around 75% less to run on average, counting a tokenizer that uses about 30% more tokens.
On Anthropic’s own table it scores 39.2% on Terminal-Bench 4.0, against 0.0% for Haiku 4.5 and 70.6% for Sonnet 5.5, and 72.4% on OSWorld 2.1 against 83.9%. Anthropic says Sonnet 5.5 and Opus 5.5 remain the better choice for complex agentic coding.
There was never a Haiku 5. Both next-week posts missed, the 1M context claim held, the 20 times cheaper claim undershot the real cut on short prompts, and the claim that it would beat Fable 5.1 is contradicted by Anthropic’s own numbers.
CellCog does not route to Haiku 5.5. Every chat tier runs on Claude Opus 5.5, with Flash at low effort. Anthropic pitches Haiku 5.5 as a subagent under an Opus or Sonnet lead, which is the pattern to watch in multi-agent harnesses.
§ 01The release, in one table
| Item | Claude Haiku 5.5 |
|---|---|
| Released | October 7, 2026 |
| Model ID | claude-haiku-5-5 |
| Context window | 1M tokens |
| Max output | 128K tokens |
| Input, prompts up to 100K | $0.10 per million tokens |
| Output, prompts up to 100K | $0.50 per million tokens |
| Input and output, prompts over 100K | $0.50 and $2.50 per million tokens |
| Cache reads | $0.01, or $0.05 over 100K |
| Thinking | Adaptive, with an effort setting (default medium) |
| Knowledge cutoff | June 2026 |
| Platforms | Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry |
It is the first Haiku with an adjustable effort setting. It also replaces a model that had been current for 357 days: Claude Haiku 4.5 shipped on October 15, 2025.
§ 02The price: a bigger cut than the rumor
The August rumors put Haiku at about 20 times cheaper than Opus 5. On output that would have been $1.25 per million tokens. The real number on short prompts is $0.50.
| Model | Output per million tokens | Source |
|---|---|---|
| Claude Haiku 5.5, prompts up to 100K | $0.50 | Anthropic pricing |
| Claude Haiku 5.5, prompts over 100K | $2.50 | Anthropic pricing |
| Claude Haiku 4.5 | $5 | Anthropic pricing |
| Claude Sonnet 5.5 | $10 | Anthropic pricing |
| Claude Opus 5.5 | $20 | Anthropic pricing |
| Claude Fable 5.1 | $50 | Anthropic pricing |
| Rumored: 20 times under Opus 5 | $1.25 | Arithmetic on the August posts |
Against Opus 5’s $25 output price, Haiku 5.5 is 50 times cheaper on short prompts and 10 times cheaper on long ones, so the 20 times rumor sat between the two. Against Opus 5.5 it is 40 times cheaper on both input and output for short prompts.
There is a catch. The model uses the newer tokenizer, so “the same text counts as approximately 30% more tokens than on Claude Haiku 4.5”. Anthropic’s own average accounts for that: “On average, it now costs around 75% less to run.” Its footnote adds that around 90% of Haiku 4.5 requests were under 100,000 tokens.
§ 03Anthropic’s benchmark table
Anthropic ran all of these numbers itself, and no independent scores exist yet. The comparison set is Haiku 4.5, OpenAI’s GPT-6 Luna, and Sonnet 5.5 for reference.
| Benchmark | Haiku 5.5 | Haiku 4.5 | GPT-6 Luna | Sonnet 5.5 |
|---|---|---|---|---|
| GDPval-AA v2.1 (score) | 1620 | 735 | 1437 | 1840 |
| AA-Briefcase v1.1 (score) | 1578 | 614 | 1336 | 1824 |
| OSWorld 2.1, offline subset | 72.4% | 15.7% | 48.9% | 83.9% |
| Humanity’s Last Exam, no tools | 45.9% | 10.2% | not reported | 56.9% |
| Humanity’s Last Exam, with tools | 57.4% | 18.7% | not reported | 64.5% |
| Terminal-Bench 4.0 | 39.2% | 0.0% | 16.4% | 70.6% |
| FrontierCode 1.1 (Main) | 46.4% | not reported | 42.4% | 52.1% (xhigh) |
| Chartography, no tools | 46.4% | 6.4% | 29.1% | 61.6% |
Haiku 5.5 beats Haiku 4.5 and GPT-6 Luna on every row where both are reported. It trails Sonnet 5.5 on every row, and the gap is widest on the long agentic coding task. Anthropic says so itself: “Sonnet 5.5 and Opus 5.5 remain better choices for complex agentic coding tasks like those measured by Terminal-Bench 4.0”.
§ 04The rumors, graded
The tracker logged nine claims. Here is each one against what shipped.
| Claim | Posted | Grade | What shipped |
|---|---|---|---|
| Haiku 5.5 follows Opus 5.5 in the coming weeks | Anthropic, Sep 22 | Confirmed | Released 15 days later |
| Haiku 5 launching next week | X, Aug 22 | Contradicted | No Haiku 5 ever shipped |
| Haiku 5 could launch next week | X, Sep 8 | Contradicted | No Haiku 5 ever shipped |
| About 20 times cheaper than Opus 5 | X, Aug 12 and 17 | Confirmed, understated | 50 times cheaper on short prompts, 10 on long |
| 1M-token context window | X, Aug and Sep | Confirmed | 1M tokens |
| Near Opus 5 performance | X, Aug 12 | Unconfirmed | Anthropic compares it with Sonnet 5.5, not Opus |
| Opus 5.5 level, beats Fable 5.1 | X, Sep 23 | Contradicted | Trails Sonnet 5.5 on every row of Anthropic’s table |
| Already finished, waiting for release | X, Sep 23 | Cannot check | No source was named |
The account that posted the Haiku 5 claims was right on the context window and right on the direction of the price. It was wrong on every date and on the claim that mattered most to buyers, that this would be a Fable-class model at Haiku prices.
§ 05What breaks if you move from Haiku 4.5
Anthropic lists these as needing code changes in What’s new in Claude Haiku 5.5:
- Manual extended thinking (
budget_tokens) returns an error; use adaptive thinking. - Non-default
temperature,top_pandtop_kreturn an error. - Assistant message prefill returns an error; end
messageswith a user turn. - Computer use needs the new
computer_toolset_20260801on the Claude API and Google Cloud. - Changing earlier turns invalidates thinking blocks, so keep conversations append-only.
- Thinking blocks work only in the account that produced them.
Responses can also begin with thinking blocks now, so select content blocks by type, not by position.
§ 06Everything else Anthropic changed on October 7
- Sonnet 5.5 cache reads halved, from $0.20 to $0.10 per million tokens. Anthropic says this makes Sonnet 5.5 around 20% cheaper on most agentic work (release notes).
- Monthly API credits for subscribers: $100 on Max 5x, $200 on Max 20x, and on Team $20 per Standard seat and $100 per Premium seat, pooled and capped at $500, usable on any model (terms). Free, Pro and Enterprise plans are not eligible, and a new subscriber waits seven days. The credits cover the API, batches, the Console Playground, Managed Agents and the Agent SDK; they do not cover interactive Claude Code, extra usage in Claude or Cowork, or Claude on Bedrock, Vertex AI or Foundry. They do not roll over, and they are spent before any credits you bought.
- Computer use and browser use in beta in the Python and TypeScript SDKs. Anthropic says Haiku 5.5 is especially well suited to them.
§ 07Anthropic’s .5 rhythm, this time
The last .5 family shipped Sonnet, then Haiku, then Opus over 56 days, starting September 29, 2025. This one shipped Opus 5.5 on September 22, Sonnet 5.5 on September 28 and Haiku 5.5 on October 7, a whole family in 15 days. Haiku came last, 9 days after Sonnet.
§ 08Where CellCog stands
CellCog does not route to Haiku 5.5. Every chat tier runs on Claude Opus 5.5 (what changed), with Flash at low effort. Anthropic’s own framing of Haiku 5.5 is a subagent under an Opus or Sonnet lead, the shape Cognition describes for Devin Fusion with Opus 5.5 as the lead. That is the pattern to watch in multi-agent harnesses. Any routing change on our side will land on this page as a dated note.
§ 09What we are watching for
- Independent speed and quality scores, starting with Artificial Analysis and the coding leaderboards.
- Whether the 75% average saving holds on real agent traffic, given the 30% tokenizer overhead.
- OpenRouter and other catalogs; at release, OpenRouter’s Anthropic list still showed Haiku 4.5 as its newest Haiku.
- A Haiku 4.5 deprecation notice.
§ 10The tracker
As of October 7, 2026, 18:17 UTC: Claude Haiku 5.5 is released, priced and documented. The Haiku 5 predicted for next week in August and September never shipped.
§ 11Update log
- As of October 8, 2026, 13:08 UTC: monthly API credit terms added from Anthropic’s help article (per-seat Team math, eligibility, what the credits cover and what they do not).
- As of October 7, 2026, 18:17 UTC: Haiku 5.5 released at 18:01 UTC. Page flipped to the release record, with specs, price, Anthropic’s benchmark table and the rumors graded.
- As of September 23, 2026, 14:40 UTC: page opened.
§ 12Sources
- Anthropic, Introducing Claude Haiku 5.5, October 7, 2026.
- Anthropic docs: Claude Haiku 5.5 overview, What’s new, pricing and release notes, read October 7, 2026.
- X: @claudeai launch thread and @AnthropicAI, October 7, 2026; times from the post ids.
- Anthropic, Introducing Claude Opus 5.5, September 22, 2026.
- X: @Mr_Salio, August 12, August 17, August 22, September 8 and September 23.
Q1When was Claude Haiku 5.5 released?
October 7, 2026. Anthropic’s model page says Released October 7, 2026, and the @claudeai launch thread is timestamped 18:01 UTC. It came 15 days after the Opus 5.5 announcement named it and 9 days after Sonnet 5.5.
Q2How much cheaper is Haiku 5.5 than Haiku 4.5?
On list price, 90% cheaper for prompts up to 100,000 tokens and 50% cheaper above. Anthropic says around 90% of Haiku 4.5 requests fell under 100,000 tokens and that the model costs around 75% less to run on average, after counting its tokenizer, which uses about 30% more tokens for the same text.
Q3Does Claude Haiku 5.5 beat Opus 5.5 or Fable 5.1?
Not on Anthropic’s numbers. It trails Sonnet 5.5 on every benchmark in Anthropic’s launch table, by 31.4 points on Terminal-Bench 4.0, and Anthropic says Sonnet 5.5 and Opus 5.5 remain better choices for complex agentic coding. The September 23 X post claiming it would beat Fable 5.1 does not hold.
Q4What breaks when moving from Haiku 4.5 to Haiku 5.5?
Manual extended thinking with budget_tokens returns an error, as do non-default temperature, top_p and top_k, and assistant prefill. Computer use needs the new toolset, responses can begin with thinking blocks, and the same text counts as about 30% more tokens. Anthropic’s migration guide covers each.
Q5What else did Anthropic change on October 7?
Sonnet 5.5 cache reads dropped from $0.20 to $0.10 per million tokens, which Anthropic says makes Sonnet 5.5 around 20% cheaper on most agentic work. Max and Team plans now include monthly API credits: $100 on Max 5x, $200 on Max 20x and up to $500 pooled on Team.