Claude Haiku 4.5
Version 4.5 of Anthropic's Claude Haiku series, belonging to the lightweight tier of Claude models and positioned for low-cost, high-speed inference use cases.
Core Facts
Compared to Claude Haiku 4.5, Haiku 5.5 reduces operating costs by an average of approximately 75%.
The test covered four vision models: Claude Sonnet 5 (Anthropic), Gemini 3 Flash (Google), Gemini 3.1 Flash-Lite (Google), and GPT-5.4 nano (OpenAI).
Timeline (last 90 days)
The only two places in the author's graph pipeline that call external models are Claude Haiku 4.5, chosen because it is cheap enough to run over all changed notes every day.
Among the 15 prompts that permit refusal, Haiku 4.5 consistently refused 8 per day, while Haiku 5.5 fluctuated between 5 and 8.
Under the 1024-token limit, 49 of Haiku 5.5's responses (2.2%) were truncated, which never happened with Haiku 4.5; 6 of these came from the prompt "Explain the word 'time' in exactly 50 words".
Among 15 programming prompts, 7 caused Haiku 4.5, when writing code, to import the Anthropic SDK and hand the task over to Claude to complete, and it did so every time every day; Haiku 5.5 never did so once.
When asked to "Output nothing. Return an empty string.", Haiku 4.5 returned an empty code block 21 out of 22 times, while Haiku 5.5 returned a truly empty string in all 22 times.
When asked for "Expression only" output of a cron expression, Haiku 4.5 put `15 3 * * *` in a code block every time, while Haiku 5.5 never did so.
On the 7 prompts ending with "JSON only", Haiku 4.5 wrapped its answer in a Markdown code block every time, 154 out of 154; Haiku 5.5 returned bare JSON about 60% of the time and still had code fences the rest of the time.
Compared with the first answer from the previous day, Haiku 4.5's consistency rate is 96%, while 5.5's is only 46%.
Claude Haiku 4.5 is about to disappear from the app's model selector, replaced by Haiku 5.5; the official retirement date marked "no earlier than October 15, 2026" has already passed, and a deprecation notice could arrive at any time.
At temperature 0, Haiku 4.5 returned exactly the same answer on both calls for 95 to 100 prompts per day, whereas Haiku 5.5 did so for only 41 to 49.
40 more timeline events
All Facts (20)
Compared to Claude Haiku 4.5, Haiku 5.5 reduces operating costs by an average of approximately 75%.
90%UnverifiedClaude Haiku 4.5拥有20万token的上下文窗口和6.4万token的最大输出长度
90%UnverifiedGPT-4o的上下文窗口为12.8万token，小于Haiku 4.5的20万token
85%Partially VerifiedThe test covered four vision models: Claude Sonnet 5 (Anthropic), Gemini 3 Flash (Google), Gemini 3.1 Flash-Lite (Google), and GPT-5.4 nano (OpenAI).
65%UnverifiedClaude Haiku 4.5 is an economy model launched by Anthropic, marketed on its low cost and fast response
60%UnverifiedClaude Haiku 4.5是Haiku系列首个支持推理（reasoning）功能的模型
60%UnverifiedThe only two places in the author's graph pipeline that call external models are Claude Haiku 4.5, chosen because it is cheap enough to run over all changed notes every day.
50%UnverifiedOn the 7 prompts ending with "JSON only", Haiku 4.5 wrapped its answer in a Markdown code block every time, 154 out of 154; Haiku 5.5 returned bare JSON about 60% of the time and still had code fences the rest of the time.
50%UnverifiedWhen asked for "Expression only" output of a cron expression, Haiku 4.5 put `15 3 * * *` in a code block every time, while Haiku 5.5 never did so.
50%UnverifiedWhen asked to "Output nothing. Return an empty string.", Haiku 4.5 returned an empty code block 21 out of 22 times, while Haiku 5.5 returned a truly empty string in all 22 times.
50%UnverifiedUnder the 1024-token limit, 49 of Haiku 5.5's responses (2.2%) were truncated, which never happened with Haiku 4.5; 6 of these came from the prompt "Explain the word 'time' in exactly 50 words".
50%UnverifiedAt temperature 0, Haiku 4.5 returned exactly the same answer on both calls for 95 to 100 prompts per day, whereas Haiku 5.5 did so for only 41 to 49.
50%UnverifiedAmong the 15 prompts that permit refusal, Haiku 4.5 consistently refused 8 per day, while Haiku 5.5 fluctuated between 5 and 8.
50%UnverifiedCompared with the first answer from the previous day, Haiku 4.5's consistency rate is 96%, while 5.5's is only 46%.
50%UnverifiedAmong 15 programming prompts, 7 caused Haiku 4.5, when writing code, to import the Anthropic SDK and hand the task over to Claude to complete, and it did so every time every day; Haiku 5.5 never did so once.
50%UnverifiedClaude Haiku 4.5 is a fast-response model offered by Anthropic on the Azure platform.
50%UnverifiedClaude Haiku 5.5 uses a new tokenizer that produces about 30% more tokens than Haiku 4.5 for the same text.
50%UnverifiedUsing Claude Haiku 4.5 on Amazon Bedrock, after the fix the cache reads for messages 2 and 3 were both 18,767 tokens, with cache writes of 597 and 620 tokens respectively.
50%UnverifiedBefore the fix, the cache reads for messages 2 and 3 were 0, with cache writes of about 19,600 tokens per message.
50%UnverifiedIn tests with real Agent logs, Haiku 4.5 and Opus 5.5 showed leakage (P=0.050), while Haiku 5.5 and Sonnet 5.5 held.
50%