The fastest Claude option at near-frontier quality
Coding quality close to Sonnet 4 at roughly one-third the cost and about 2x the speed; computer use can exceed Sonnet 4.
Overview
Claude Haiku 4.5 is Anthropic's fast, low-latency Claude model released on October 15, 2025. Anthropic positions it as the fastest near-frontier option: coding quality close to Claude Sonnet 4 at roughly one-third the cost and about 2x the speed, with computer-use results that can exceed Sonnet 4.
That makes Haiku 4.5 a strong default for real-time chat, support agents, pair-programming loops, and multi-agent systems where many subtasks run in parallel. It also supports Extended Thinking, so teams can spend a little more reasoning budget on harder steps without moving every request to Sonnet or Opus.
Key capabilities
| Dimension | Detail |
|---|---|
| Context window | 200,000 tokens |
| Max output | 65,536 tokens |
| Input modalities | Text, image |
| Output modalities | Text |
| Tools | function calling, structured outputs, streaming, computer use |
| Reasoning | Extended Thinking |
Prompt caching uses product-level multipliers: cache reads are 0.1x the base input rate, 5-minute writes are 1.25x, and 1-hour writes are 2x. See live pricing in the model catalog.
Benchmarks
Haiku 4.5's story is price-performance: it delivers near-Sonnet coding at the lowest latency in the family.
Software engineering: SWE-bench Verified
On SWE-bench Verified (n=500), Haiku 4.5 scores 73.3%, slightly ahead of Sonnet 4 at 72.7% and close to GPT-5 Codex at 74.5%. That is unusually strong for a latency-oriented model and makes it attractive for IDE assistants, small fixes, test generation, and code-review triage.
Speed and cost positioning
The best balance of speed and unit cost
SWE-bench Verified coding edges out Sonnet 4 while keeping the lowest latency and highest throughput in the Claude family.
Haiku 4.5 runs roughly 4-5x faster than Sonnet 4.5 while retaining about 90% of its agentic coding performance — the basis for Anthropic's recommendation to use it as a parallel worker under a Sonnet or Opus orchestrator. Its safety profile is also favorable: misaligned behavior is significantly lower than Sonnet 4.5 and Opus 4.1, and the model is classified as ASL-2.
When to use it
- Realtime support and chat agents: low latency with strong language quality.
- Pair programming and IDE plugins: fast output plus a 73.3% SWE-bench Verified score.
- High-volume text processing: summarization, classification, extraction, and rewriting.
- Sub-agents: run many Haiku workers under a Sonnet or Opus orchestrator to reduce total cost.
CrossModel exposes Claude Haiku 4.5 through Anthropic-compatible /v1/messages and OpenAI-compatible /v1/chat/completions. Current pricing is available in the model catalog.