Claude Haiku 4.5
2025-10-15 · Publisher : Anthropic · Domain : Artificial Intelligence
Anthropic launches Claude Haiku 4.5, a fast model that delivers coding performance close to Sonnet 4 at one third of the cost and more than twice the speed, under ASL-2.
What changes
- Coding performance similar to Sonnet 4 at one third of the cost and more than twice the speed.
- Outperforms Sonnet 4 on some tasks, such as computer use.
- About 90% of Sonnet 4.5's performance in agentic coding, 4 to 5 times faster.
- Released under AI Safety Level 2, with a rate of misaligned behavior statistically lower than Sonnet 4.5 and Opus 4.1.
- Suited to sub-agents and real-time applications.
### Specifications - API identifier: claude-haiku-4-5-20251001 (alias claude-haiku-4-5). - 200K-token context, maximum output 64K tokens. Reliable knowledge cutoff: February 2025; training data up to July 2025. - Manual extended thinking. Text and images as input, text as output. - Availability: Claude.ai, Claude Code, API, Amazon Bedrock, Google Cloud Vertex AI. ### Pricing | Input (USD / M tokens) | Output (USD / M tokens) | Cache write 5 min / 1 h | Cache read | Batch | |---|---|---|---|---| | 1 | 5 | 1.25 / 2 | 0.10 | 0.50 / 2.50 | ### Benchmarks | Benchmark | Score | |---|---| | SWE-bench Verified (average of 50 trials) | 73.3% | | Terminal-Bench (without thinking) | 40.21% | | Terminal-Bench (with 32K thinking) | 41.75% | Scores published by the vendor; the full comparison table (Sonnet 4.5, Opus 4.1, GPT-5, Gemini) is an image, not recorded. ### Key takeaways Haiku 4.5 sharply lowers the cost of assisted coding and sub-agents, with a level close to the frontier of five months earlier.
Links
Tags : LLM, Anthropic, Claude, Haiku, Faible latence, Codage