Anthropic releases Claude Haiku 5.5 in the Claude API
Claude Haiku 5.5 is now available in the Claude API with a 1M token context window and up to 128k output tokens.
On Zabio today
claude-haiku-5-5Live. Call it by name with your Zabio key.Price and detailsclaude-haiku-4-5Live. Call it by name, or let auto pick it when it fits the job.Price and details
What changed
Anthropic launched Claude Haiku 5.5 (claude-haiku-5-5), its most capable model for high-volume and latency-sensitive work. It has a 1M token context window, up to 128k output tokens, and adaptive thinking with the effort parameter. It is available on the Claude API, Claude in Amazon Bedrock, Claude Platform on AWS, Claude on Google Cloud, and Claude in Microsoft Foundry. Compared with Claude Haiku 4.5, the same text counts as approximately 30% more tokens.
Key details
- Context window of 1M tokens.
- Up to 128k output tokens.
- Adaptive thinking with the effort parameter.
- Available on the Claude API, Claude in Amazon Bedrock, Claude Platform on AWS, Claude on Google Cloud, and Claude in Microsoft Foundry.
- Prompts over 100,000 tokens cost more.
- Minimum cacheable prompt length is 512 tokens, down from 4,096 on Claude Haiku 4.5.
What it means for Zabio customers
Token counts, max_tokens values and cost estimates need to be rechecked. Thinking tokens count toward max_tokens, so a small limit can stop after a thinking block and before any text.
Breaking changes and migration
Code written for Claude Haiku 4.5 can break. Replace budget_tokens with adaptive thinking, because manual extended thinking returns a 400 error. Select content blocks by type, because responses can begin with thinking blocks.
Source
We’ve launched Claude Haiku 5.5 (claude-haiku-5-5), our most capable model tuned for high-volume and latency-sensitive work. It has a 1M token context window, 128k max output tokens, and adaptive thinking with the effort parameter.
Original announcement: Claude Platform release notes
How it compares
The models in this story next to models of the same class at a similar price, all available with one Zabio key.
| Model | Input, $/M | Output, $/M | Cached, $/M | Context window | 1,000 calls | Fixing code, hard |
|---|---|---|---|---|---|---|
claude-haiku-5-5AnthropicIn this story | 0.10 | 0.50 | 0.01 | 1,000,000 | $0.45 | 77.6 |
claude-haiku-4-5AnthropicIn this story | 1.00 | 5.00 | 0.10 | 200,000 | $4.50 | 13.9 |
gpt-6-lunaOpenAI | 0.10 | 0.50 | 0.01 | 400,000 | $0.45 | 70.8 |
llama-guard-4-12bMeta | 0.18 | 0.18 | no data | 131,072 | $0.45 | no data |
qwen3.5-9bQwen | 0.17 | 0.25 | no data | 262,144 | $0.47 | 9.2 |
Prices are the vendor list prices in USD per million tokens. "1,000 calls" is 1,000 requests of 2,000 input tokens and 500 output tokens each, without caching. Zabio bills the list price plus its service fee: see pricing.
Scores: percent correct for that kind of job, aggregated from public results by Epoch AI and LiveBench, read on 16 November 2026. Epoch AI: CC BY 4.0, LiveBench: Leaderboard results used with attribution; the 2.0 release states no terms of its own (read 2026-11-06); the first version's Mathematics, Language, IF results: Code Apache 2.0 and MIT; leaderboard results used with attribution
Source: Claude Platform release notes (2026-10-07)