Claude Haiku 5.5: model ids, price and cutoff

How to call it

6 ways to use Claude Haiku 5.5, with the model id each provider expects.

Providers, model ids, endpoints and docs for Claude Haiku 5.5
Provider	Model id	Endpoint	Docs
Anthropic API (Claude API)	`claude-haiku-5-5`	`https://api.anthropic.com/v1/messages`	platform.claude.com
AWS Bedrock	`anthropic.claude-haiku-5-5`		platform.claude.com
Google Cloud Vertex AI	`claude-haiku-5-5`		platform.claude.com
Microsoft Foundry (Azure)	`claude-haiku-5-5`		platform.claude.com
Claude Platform on AWS	`claude-haiku-5-5`		platform.claude.com
Web app			claude.ai

Pricing

Prices for Claude Haiku 5.5
Input	$0.10 per 1M tokens
Output	$0.50 per 1M tokens
Cache read	$0.01 per 1M tokens

As published: “per 1M tokens (USD) for prompts up to 100K tokens; above 100K: $0.50 input / $2.50 output / $0.05 cache read; 5-min cache write $0.125 ($0.625 >100K); Batch API 50% off”.

Source: platform.claude.com, checked 8 October 2026.

Specs

Input: Text and images
Output: Text
Context window: 1,000,000 tokens
Max output: 128,000 tokens
Open weights: No
Licence: Proprietary
Training cutoff: June 2026
Released: 7 October 2026
Status: Current
Type: Reasoning model

What stands out

What its maker marketed at launch, and what people found later, each with its source.

First Haiku with effort control

First

Adaptive thinking steered by the effort parameter (low, medium, high, xhigh, max; default medium); thinking: {type: disabled} works at high effort or below and returns 400 at xhigh/max; per-message effort changes in beta.

Source: platform.claude.com
Two-tier long-context pricing

$0.10/$0.50 per MTok up to 100K prompt tokens, five times that above 100K; 90% below Haiku 4.5 and equal to GPT-6 Luna at the low tier.

Source: platform.claude.com
Frontier-level prompt-injection robustness in a small model

System card: Gray Swan IPI attack success at k=15 of 7.1% (Haiku 4.5: 83.2%); 0.08% against the Shade adaptive coding attacker (Haiku 4.5: 58.40%).

Source: anthropic.com
Strong computer use for its price

Anthropic reports OSWorld 2.1 at 72.4%, versus 48.9% for GPT-6 Luna.

Source: anthropic.com

Four assistants, one scale

Claude Haiku 5.5 is added as a fifth row. Each row runs from what the model knows (graphite) through the months between its cutoff and its launch (hatched) to today (pink).

GPT-6 Astra 228 at launch, 771 today
Claude Opus 5.5 305 at launch, 711 today
Gemini 3.8 Flash 246 at launch, 794 today
Grok 4.7 317 at launch, 740 today
Claude Haiku 5.5 671 at launch, 711 today

Show as a table

Model	Cutoff	Released	Missing at launch	Missing today
GPT-6 Astra	April 2026	3 September 2026	228	771
Claude Opus 5.5	June 2026	22 September 2026	305	711
Gemini 3.8 Flash	March 2026	2 September 2026	246	794
Grok 4.7	May 2026	21 September 2026	317	740
Claude Haiku 5.5	June 2026	7 October 2026	671	711

Notes

Anthropic’s fastest and cheapest current model, for classification, extraction, routing and subagents. Uses the Claude 4.7+ tokenizer (~30% more tokens than Haiku 4.5 for the same text). Migrating from Haiku 4.5: manual extended thinking (budget_tokens) returns 400; responses can begin with thinking blocks. Non-default temperature/top_p/top_k return 400. Batch API supports 300K output with the output-300k-2026-03-24 beta header. Retirement not sooner than 2027-10-07.

curl https://api.anthropic.com/v1/messages \
 -H "x-api-key: $ANTHROPIC_API_KEY" -H "anthropic-version: 2023-06-01" -H "content-type: application/json" \
 -d '{"model":"claude-haiku-5-5","max_tokens":4000,"output_config":{"effort":"low"},"messages":[{"role":"user","content":"Classify: ..."}]}'

Sources: