Claude Haiku 4.5
Claude Haiku 4.5
Anthropic's fast-tier model for the Claude 4 generation. Released October 15, 2025. Replaces claude-3-5-haiku and functions as a cost-efficient alternative to claude-sonnet-4-6 for high-volume workloads. First Haiku model to support extended-thinking and computer-use.
Specs
| Property | Value |
|---|---|
| API model ID | claude-haiku-4-5-20251001 |
| Context window | 200,000 tokens |
| Max output | 64,000 tokens |
| Knowledge cutoff | February 2025 |
| Input modalities | Text, images |
Benchmarks
| Benchmark | Score |
|---|---|
| SWE-bench Verified | 73.3% |
Matches claude-sonnet-4-6 on coding, computer-use, and agent tasks. Roughly tied with GPT-5 on SWE-bench Verified at approximately one-third the price of Sonnet 4.5.
Pricing
| Direction | Cost |
|---|---|
| Input | $1.00 / MTok |
| Output | $5.00 / MTok |
One-third the price of claude-sonnet-4-6 ($3/$15). Slightly above claude-3-5-haiku ($0.80/$4) but with substantially expanded capabilities.
Speed
More than 2× faster than claude-sonnet-4-6. Time to first token approximately 300ms. Throughput roughly 226 characters/second via Vertex AI, 136 c/s via Anthropic direct.
Capabilities
- extended-thinking: Supported (evaluated with 128K thinking budget). First Haiku model to include it.
- computer-use: Supported. Outperforms claude-sonnet-4-6 on certain computer interaction tasks.
- Vision: Text and image inputs.
Use Cases
Best for: sub-agent roles in agentic-workflows (e.g., Sonnet 4.5 orchestrates multiple Haiku instances), high-volume classification and extraction, real-time chat and customer service, pair programming, low-latency completions.
Not ideal for: complex long-horizon reasoning where spending on Sonnet or Opus tier is justified.
Availability
Claude API (claude-haiku-4-5-20251001), AWS Bedrock, Google Cloud Vertex AI, claude.ai (default free-tier model).
Related
claude-sonnet-4-6 · claude-opus-4-6 · extended-thinking · computer-use · agentic-workflows · evals