Claude Haiku 4.5
Anthropic's fastest, cheapest model with near-frontier intelligence.
Context window
200K
Input / 1M tokens
$1.00
Output / 1M tokens
$5.00
Provider
Anthropic
Pricing $1/$5 per 1M input/output tokens, with up to 90% savings via prompt caching and 50% via batch processing. Lowest-cost current Claude model. · Data verified 2026-07-02
Claude Haiku 4.5 (October 2025) is Anthropic's fastest and most efficient model, delivering near-frontier intelligence at a fraction of the cost and latency of larger Claude models - matching roughly Sonnet 4-level performance on reasoning, coding, and computer use. It scored 73.3% on SWE-bench Verified and is the first Haiku with extended thinking, computer use, and context awareness. It has a 200K context window and 64K max output.
Capability index
Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.
How to access it
Available via the Claude API (alias claude-haiku-4-5, pinned snapshot claude-haiku-4-5-20251001), Amazon Bedrock, Google Cloud, and Microsoft Foundry, and on claude.ai. Ideal as a fast sub-agent or high-throughput worker.
Strengths
- ✓Fastest latency and lowest price ($1/$5) in the current Claude lineup
- ✓Near-frontier coding for its class: 73.3% SWE-bench Verified
- ✓First Haiku with extended thinking, computer use, and context awareness
- ✓Excellent for sub-agents, parallelized execution, and scaled deployment
- ✓Handles both text and image input
Best for developers who...
When to choose it (and when not to)
Reach for Claude Haiku 4.5 when...
- →You need high throughput and low latency at minimal cost
- →You are fanning out many parallel sub-agents or worker calls
- →You want solid coding/computer-use quality without Opus/Sonnet pricing
- →Your task fits comfortably within a 200K context window
Look elsewhere if...
- ✕You need frontier-level reasoning or the hardest coding tasks (use Sonnet 5 or Opus 4.8)
- ✕You need more than 200K context or more than 64K output tokens
- ✕You need Jan-2026-era knowledge (Haiku 4.5's reliable cutoff is Feb 2025)
How to use it
- ›Turn on extended thinking for harder reasoning/coding steps; leave it off for simple, latency-critical calls
- ›Use it as a fast sub-agent under an Opus/Sonnet orchestrator for cost-efficient fan-out
- ›Keep prompts tight and within 200K context; cache shared prefixes for repeated calls
- ›Batch non-interactive jobs for 50% savings
Quickstart
Pythonimport anthropic
client = anthropic.Anthropic()
response = client.messages.create(
model="claude-haiku-4-5",
max_tokens=2048,
messages=[
{"role": "user", "content": "Classify these support tickets by urgency."}
],
)
print(response.content[0].text)Alias claude-haiku-4-5 resolves to snapshot claude-haiku-4-5-20251001. Supports extended thinking via the thinking parameter.
API model id: claude-haiku-4-5-20251001
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| SWE-bench Verified | 73.3% | Averaged over 50 trials with 128K thinking budget (Anthropic) |
| GPQA | 64.6% | Per Artificial Analysis |
| MMLU Pro | 80.0% | Per Artificial Analysis |
Tools powered by Claude Haiku 4.5
Compare Claude Haiku 4.5 with
Claude Haiku 4.5 vs Claude Sonnet 4.6
Anthropic - 1M ctx
Claude Haiku 4.5 vs Gemini 2.5 Flash
Google DeepMind - 1,048,576 tokens (1M) input; up to 65,535 output ctx
Claude Haiku 4.5 vs GPT-4o
OpenAI - 128,000 tokens (16,384 max output) ctx