Claude Haiku 4.5

What is disclosed

Claude Haiku 4.5 was released on 15 October 2025.¹¹1Anthropic, Introducing Claude Haiku 4.5, 15 October 2025. https://www.anthropic.com/news/claude-haiku-4-5 Anthropic calls it a small model. Nothing on its size or architecture has been published.

Its context window is 200K tokens, and a response stops at 64K.²²2Anthropic, Claude Haiku 4.5 model page, Claude Platform docs, read 29 September 2026. https://platform.claude.com/docs/en/models/haiku-4-5/overview It accepts images as well as text. Its training data runs to July 2025, but Anthropic calls its knowledge reliable only up to February 2025.²²

Thinking is manual. A request switches it on and gives it a budget in tokens.²² There is no effort parameter, and no other current Claude model works this way.²²

It shipped under Anthropic's AI Safety Level 2 standard, ASL-2. Sonnet 4.5 and Opus 4.1 were under the more restrictive ASL-3.¹¹

What is new

At launch Anthropic set it against Claude Sonnet 4. It claimed similar coding performance "at one-third the cost and more than twice the speed".¹¹ The post also proposed a division of labour, with Sonnet 4.5 planning a complex problem and several Haiku 4.5 instances working through the subtasks in parallel.¹¹

Input costs $1 per million tokens and output $5.³³3Anthropic, Claude API pricing, read 29 September 2026: Haiku 4.5 at $1 input, $5 output and $0.10 for cache reads; Haiku 3.5 at $0.80 and $4. https://platform.claude.com/docs/en/about-claude/pricing. Arithmetic: 1 / 0.80 = 5 / 4 = 1.25. Cache reads cost $0.10 per million.³³ The previous Haiku, Claude Haiku 3.5, is priced at $0.80 and $4. Haiku 4.5 costs 25% more per token.³³

Anthropic reports 73.3% on SWE-bench Verified, averaged over 50 trials with a 128K thinking budget. The model worked with a bash tool and a file editor.¹¹ The prompt asked it to use tools "as much as possible, ideally more than 100 times".¹¹ On Terminal-Bench it averaged 40.21% with thinking off and 41.75% with a 32K budget.¹¹

Anthropic's automated alignment assessment found a lower overall rate of misaligned behavior than in Sonnet 4.5 and Opus 4.1.¹¹ By that one measure, Anthropic called it its safest model yet.¹¹

Using it

The pinned id is claude-haiku-4-5-20251001, with claude-haiku-4-5 as an alias.²² On Google Cloud the date follows an at sign, as claude-haiku-4-5@20251001.²² Batch requests are billed at half price.²²

Anthropic guarantees availability only until 15 October 2026.²² The Sonnet 5.5 launch post says a Claude Haiku 5.5 will join the family "in the coming weeks", with no date.⁴⁴4Anthropic, Introducing Claude Sonnet 5.5, 28 September 2026. https://www.anthropic.com/claude-sonnet-5-5

Claude Sonnet 5.5 sits above it in the lineup, at twice the price per token and five times the context.⁵⁵5Anthropic, Claude API pricing and Claude Sonnet 5.5 model page, read 29 September 2026: Sonnet 5.5 at $2 and $10 with a 1M-token context window. https://platform.claude.com/docs/en/about-claude/pricing and https://platform.claude.com/docs/en/models/sonnet-5-5/overview. Arithmetic: 2 / 1 = 10 / 5 = 2; 1,000,000 / 200,000 = 5.