Claude Haiku 5.5 (Oct 7, 2026): $0.10/$0.50 under 100K tokens, Haiku 4.5 breaks, Claude Code 2.1.293

On Oct 7, 2026, Anthropic launched Claude Haiku 5.5 (claude-haiku-5-5). Anthropic's pricing doc (checked 2026-10-08) lists $0.10 input / $0.50 output per million tokens for prompts up to 100,000 tokens, and $0.50 / $2.50 for prompts over that. It has a 1M-token context window, 128K max output, adaptive thinking and the effort parameter. Code written for Haiku 4.5 can break: budget_tokens returns a 400, thinking is on by default, and the same text counts as more tokens. Claude Code 2.1.293 makes it the default Haiku, on the Anthropic API only.

Price, and the 100K-token split

Source: Anthropic pricing doc, accessed 2026-10-08.

Other recent Claude models bill the whole 1M window at one rate. Haiku 5.5 is the exception: "Claude Haiku 5.5 is priced by prompt length: a prompt of over 100,000 tokens pays higher prices." The split is per request: a long prompt pays five times the per-token rate of a short one.

Haiku 4.5 lists at $1 / $5. Anthropic says Haiku 5.5 "costs around 75% less to run" on average. That is Anthropic's estimate, not ours. Its footnote: 90% cheaper than Haiku 4.5 up to 100K tokens, 50% cheaper above, about 90% of Haiku 4.5 requests under the threshold, and a new tokenizer that "uses slightly more tokens per task." Your saving depends on your prompt-length mix.

Bedrock and Google Cloud publish their own rates; Claude Platform on AWS and Microsoft Foundry bill at standard Claude API rates (pricing doc). Plan context: Claude pricing.

When to use Haiku 5.5, and when not to

When to use: Anthropic positions it for high-volume, cost-sensitive work (summaries, compactions, database queries, classification) and speed-sensitive jobs like live support and browser use. It "pairs well with Opus 5.5 and Sonnet 5.5 as a subagent on coding work." See Claude Code subagents.

When not to: Anthropic says Claude Sonnet 5.5 and Claude Opus 5.5 "remain better choices for complex agentic coding tasks." One data point from the launch page: Terminal-Bench 4.0 at 39.2% for Haiku 5.5 vs 70.6% for Sonnet 5.5 (Anthropic's numbers).

Specs (models overview): 1M context · 128K max output · default effort medium · reliable knowledge cutoff Jun 2026 · retirement not sooner than Oct 7, 2027. Anthropic calls it the first Haiku-class model with an adjustable effort setting (effort levels in Claude Code).

Available on the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry (release notes). The ID is claude-haiku-5-5 everywhere except Bedrock (anthropic.claude-haiku-5-5).

What breaks from Haiku 4.5

Anthropic's Oct 7, 2026 release notes, verbatim:

Code written for Claude Haiku 4.5 can break on Claude Haiku 5.5. Manual extended thinking (budget_tokens) returns a 400 error, and adaptive thinking is on by default, so a response can begin with thinking blocks. The same text also counts as more tokens.

In practice:

1. Remove budget_tokens before you switch the model ID, or the request fails with a 400.

2. Handle leading thinking blocks. Code that treats the first content block as the answer text will break.

3. Re-check token budgets. Limits and cost estimates tuned on Haiku 4.5 will drift. A Haiku-specific inflation figure is not published as of 2026-10-08; measure your own prompts with the token counting endpoint.

For the full checklist, follow Anthropic's Haiku 5.5 migration guide.

Claude Code 2.1.293: Haiku 5.5 becomes the default Haiku

The CHANGELOG entry for 2.1.293, verbatim:

Added Claude Haiku 5.5 (claude-haiku-5-5), now the default Haiku model on the Anthropic API — 1M context, $0.10/$0.50 per Mtok ($0.50/$2.50 for prompts over 100K)

"On the Anthropic API" matters. Where the haiku alias points (model-config docs):

Workflow

• Run claude update. The docs say: "Use v2.1.293 or later with Haiku 5.5."
• Pick it explicitly with /model claude-haiku-5-5, or start with claude --model claude-haiku-5-5.
• To pin a specific Haiku, set ANTHROPIC_DEFAULT_HAIKU_MODEL to the full model name.

Pitfalls

• On the Anthropic API, Haiku 5.5 gets the 1M window on every plan (no [1m] suffix) and auto-compacts at about 967K tokens. Long sessions cross 100K fast, and every request over 100K pays the higher rate. Prompt caching helps.
• You can't turn thinking off on Haiku 5.5 in Claude Code; /config shows "Thinking can't be turned off."
• If your model setting is haiku, a session saved on Haiku 4.5 resumes on Haiku 5.5.

Also launched Oct 7

• Sonnet 5.5 cache reads dropped from $0.20 to $0.10 per million tokens: "Cache writes and all other prices are unchanged" (release notes). Anthropic says that 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, up to $500 pooled on Team (Anthropic's credits doc).

FAQ

How much does Claude Haiku 5.5 cost?

$0.10 / $0.50 per million tokens up to 100K-token prompts; $0.50 / $2.50 above (pricing doc, 2026-10-08).

What breaks when moving from Haiku 4.5 to 5.5?

budget_tokens returns a 400, responses can start with thinking blocks, and the same text counts as more tokens.

Does Claude Code use Haiku 5.5 by default?

Yes on the Anthropic API from 2.1.293. On Claude Platform on AWS, Bedrock, Google Cloud and Foundry, haiku still resolves to Haiku 4.5.

Can I turn thinking off?

Not in Claude Code. On the API, adaptive thinking is on by default; the migration guide covers request options.

Sources (accessed 2026-10-08, ~08:13 CEST)

• Anthropic: Claude Haiku 5.5 (Oct 7, 2026): positioning, ~75% claim and footnote, Terminal-Bench line, Sonnet 5.5 cache cut, API credits
• Claude Platform release notes (Oct 7, 2026): launch, specs, platforms, Haiku 4.5 breaks
• Pricing: Haiku 5.5 rates, batch, long-context exception, cloud billing
• Models overview: IDs, default effort, cutoff, retirement
• Claude Code CHANGELOG: entry 2.1.293
• Claude Code model configuration: haiku alias per provider, 967K auto-compact, thinking, resume