CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins90% CONFIDENCEOverall confidence: 90%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better
Claude's[2][3] release notes: "November 4th, 2024 Claude Haiku 3.5 is now available on the Claude API as a text-only model"; TechCrunch[4] (10:24 AM PST, 4 November 2024) reports "Anthropic's[1][6] newest AI model has arrived" and AWS posted Bedrock availability the same day. The 22 October post had said it "will be released later this month".- Start
- Nov 4, 202492% CONFIDENCEplatform.claude.com
Claude 3.5 Haiku Released
Anthropic releases Claude 3.5 Haiku on its API, Amazon Bedrock and Vertex AI, a text-only small model that it says beats Claude 3 Opus on many benchmarks, at $1 per million input tokens, four times Claude 3 Haiku's price.
- Anthropic announced Claude 3.5 Haiku on 22 October 2024 alongside the upgraded Claude 3.5 Sonnet and computer use, and released it on 4 November on the Claude API, Amazon Bedrock and Google Cloud's Vertex AI[2][5][6]
- It launched at $1 per million input tokens and $5 per million output tokens, four times Claude 3 Haiku's price, after Anthropic said that in final testing it had "surpassed Claude 3 Opus, our previous flagship model, on many benchmarks"[4]
- It shipped text-only; Claude 3 Haiku stayed available for users needing "maximum cost-efficiency and image processing", and vision support came on 24 February 2025[2][4]
- Anthropic cut the price to $0.80 and $4 per million tokens in an update dated 3 December 2024[6]
- Deprecated on 19 December 2025 and retired from the Claude API on 19 February 2026, replaced by Claude Haiku 4.5[3]
Notable features
- The next generation of Anthropic's fastest model: at a speed similar to Claude 3 Haiku it "improves across every skill set" and in many cases matches the original Claude 3.5 Sonnet and Claude 3 Opus[1][6]
- Coding: 40.6% on SWE-bench Verified, above the original Claude 3.5 Sonnet's 33.4% and Claude 3 Opus's 22.2%, and 74% on Anthropic's internal agentic coding test against Opus's 38%[1][6]
- Other reported scores: 88.1% on HumanEval, 41.6% on GPQA Diamond, 65.0% on MMLU Pro and 51.0% on TAU-bench retail against Claude 3 Haiku's 75.9%, 33.3%, 49.0% and 18.2%[1]
- A July 2024 knowledge cutoff, more recent than Claude 3 Haiku's, and longer outputs; pitched for user-facing products, sub-agent tasks and personalising from large datasets[1][4][6]
Benchmarks[6]
| Benchmark | Claude 3.5 Sonnet (new) | Claude 3.5 Haiku | Claude 3.5 Sonnet | GPT-4o* | GPT-4o mini* | Gemini 1.5 Pro | Gemini 1.5 Flash |
|---|---|---|---|---|---|---|---|
| GPQA (Diamond) (graduate level reasoning) | 65.0% 0-shot CoT | 41.6% 0-shot CoT | 59.4% 0-shot CoT | 53.6% 0-shot CoT | 40.2% 0-shot CoT | 59.1% 0-shot CoT | 51.0% 0-shot CoT |
| MMLU Pro (undergraduate level knowledge) | 78.0% 0-shot CoT | 65.0% 0-shot CoT | 75.1% 0-shot CoT | — | — | 75.8% 0-shot CoT | 67.3% 0-shot CoT |
| HumanEval (code) | 93.7% 0-shot | 88.1% 0-shot | 92.0% 0-shot | 90.2% 0-shot | 87.2% 0-shot | — | — |
| MATH (math problem-solving) | 78.3% 0-shot CoT | 69.2% 0-shot CoT | 71.1% 0-shot CoT | 76.6% 0-shot CoT | 70.2% 0-shot CoT | 86.5% 4-shot CoT | 77.9% 4-shot CoT |
| AIME 2024 (high school math competition) | 16.0% 0-shot CoT | 5.3% 0-shot CoT | 9.6% 0-shot CoT | 9.3% 0-shot CoT | — | — | — |
| MMMU (visual Q/A) | 70.4% 0-shot CoT | — | 68.3% 0-shot CoT | 69.1% 0-shot CoT | 59.4% 0-shot CoT | 65.9% 0-shot CoT | 62.3% 0-shot CoT |
| SWE-bench Verified (agentic coding) | 49.0% | 40.6% | 33.4% | — | — | — | — |
| TAU-bench (agentic tool use) | Retail 69.2%, Airline 46.0% | Retail 51.0%, Airline 22.8% | Retail 62.6%, Airline 36.0% | — | — | — | — |
* Anthropic's evaluation tables exclude OpenAI's o1 model family, which depends on extensive pre-response computation time and makes comparisons difficult.
Buy onClaude