ReviewApp reviewSource: Anthropic
Review: Claude Haiku 5.5, a small model for high-volume tasks
Claude Haiku 5.5 offers low API rates for summaries, classification and narrowly scoped coding tasks, but input and output rates rise fivefold above 100,000 prompt tokens. Anthropic launched the model on October 7, 2026, with text-and-image input and text output. According to its documentation, context capacity increases from Haiku 4.5’s 200,000 tokens to one million, while maximum output doubles to 128,000 tokens. Adaptive thinking replaces manual thinking budgets, with medium effort as the API default.
AI language model
Claude Haiku 5.5
VerdictWorth trying
Claude Haiku 5.5 is worth evaluating for high-volume, narrowly scoped tasks where developers can verify results and measure total costs.
- Best for
- Developers building classification, summarization, context compaction or narrowly scoped coding subagents, especially with prompts up to 100,000 tokens
- Pricing
- Claude API, USD per million tokens: $0.10 input/$0.50 output for prompts up to 100,000 tokens; $0.50 input/$2.50 output above that threshold. Batch processing discounts input and output by 50%.
Pros
- Low published API rates for prompts up to 100,000 tokens
- One-million-token context and 128,000-token maximum output
- Prompt caching available from 512 tokens
- Lower costs in DataCamp’s small invoice-classification test
Cons
- Fivefold input and output rates above 100,000 prompt tokens
- Approximately 30% more input tokens with the newer tokenizer, according to Anthropic
- Incompatible manual thinking budgets and assistant prefills
- Substantial gap to Sonnet 5.5 in vendor-reported coding benchmarks
- Structured outputs unavailable on Amazon Bedrock
Why it matters
Haiku 5.5 costs less than Sonnet 5.5: published input/output rates per million tokens are $0.10/$0.50 for prompts up to 100,000 tokens, versus Sonnet’s $2/$10. DataCamp found both Haiku versions correctly classified all 24 invoice-payment decisions, with 5.5 costing less, though different thinking settings limit the comparison. Anthropic’s coding benchmarks still show a substantial gap to Sonnet, and the company recommends larger models for complex coding agents. Long-prompt rates, changed tokenization and verification costs mean developers should compare cost per accepted result on their own workloads.
Attribution
Sources
- AnthropicClaude Haiku 5.5 - Claude Platform Docs
- AnthropicClaude Haiku 5.5 migration guide - Claude Platform Docs
- AnthropicIntroducing Claude Haiku 5.5
- AnthropicWhat's new in Claude Haiku 5.5 - Claude Platform Docs
- DataCampClaude Haiku 5.5: Features, Benchmarks, and Pricing
- AnthropicModels overview - Claude Platform Docs
- AnthropicClaude
Read the original source for full details and context.