Model Details
Claude Haiku 4.5
Quick answer: Claude Haiku 4.5 (released October 15, 2025) is Anthropic's fastest and most cost-efficient model, scoring 73.3% on SWE-bench Verified — one of the world's highest scores — at $1/$5 per million tokens and running 2×+ faster than Sonnet 4.5. It brings Sonnet 4-level coding performance to the Haiku price tier.
At a Glance
Where Claude Haiku 4.5 leads
- 73.3% SWE-bench Verified at $1/$5 — frontier coding at sub-agent pricing
- 2×+ faster than Sonnet 4.5 — the fastest response time in the Claude 4 series
- Anthropic's safest model by automated alignment assessment (lower misaligned behavior rate than Sonnet 4.5 and Opus 4.1)
- ASL-2 classification — fewer restrictions, simpler deployment for most enterprise use cases
Where it lags
- Extended thinking available but with a smaller thinking budget vs Opus/Sonnet class
- Not the top choice for the most complex, open-ended reasoning that requires Opus-level depth
Best for: high-volume sub-agents, real-time chat, coding assistants, parallel agent swarms, and any workload where cost-per-call and latency are primary constraints.
What Claude Haiku 4.5 Is
Claude Haiku 4.5 was released on October 15, 2025. Anthropic's framing was direct: "What was recently at the frontier is now cheaper and faster." Five months earlier, Claude Sonnet 4 was the state-of-the-art coding model. Haiku 4.5 delivers equivalent coding performance at one-third the cost and more than twice the speed. GitHub Copilot confirmed it matches Sonnet 4 quality in code generation with faster responsiveness. Augment Code's evaluation put it at 90% of Sonnet 4.5's performance — a figure that would have been state-of-the-art just months prior.
From a Benchgen perspective, Haiku 4.5 resets the economics of multi-agent systems. A Sonnet-class orchestrator running dozens of parallel Haiku 4.5 sub-agents is now both cheaper and faster than a single Sonnet model doing sequential work.
Specifications
| Field | Value |
|---|---|
| Organization | Anthropic |
| API identifier | claude-haiku-4-5 |
| Context window | 200,000 tokens |
| License | Proprietary |
| Release date | October 15, 2025 |
| Modality | Multimodal (text and vision) |
| Safety level | ASL-2 |
Pricing
| Input (per 1M tokens) | Output (per 1M tokens) | |
|---|---|---|
| Anthropic | $1.00 | $5.00 |
Prompt caching up to 90% savings; batch processing 50% savings. Source: Anthropic pricing.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| SWE-bench Verified | 73.3% | Anthropic — Introducing Claude Haiku 4.5 | 2025-10 |
| Terminal-bench 2.0 | 40.9% avg | Anthropic — Introducing Claude Haiku 4.5 | 2025-10 |
Scores reported by Anthropic. Not Benchgen measurements. SWE-bench score averaged over 50 trials with 128K thinking budget.
Claude Haiku 4.5 vs Alternatives
| Model | SWE-bench Verified | Price (in/out per 1M) | Speed |
|---|---|---|---|
| Claude Haiku 4.5 | 73.3% | $1 / $5 | Fastest in Claude 4 |
| Claude Sonnet 4 | 72.7% | $3 / $15 | Moderate |
| Claude Opus 4.6 | 81.42% | $5 / $25 | Standard |
| Claude 3.5 Haiku | 40.6% | $1 / $5 | Fast |
Haiku 4.5 is now faster and cheaper than Sonnet 4 while matching or exceeding it on SWE-bench. The only reasons to move up to Sonnet 4.6 are tasks requiring longer outputs, deeper reasoning, or the 1M context window. (Scores from Anthropic announcements; not Benchgen measurements.)
Related Models & Benchmarks
- Compare: Claude Sonnet 4.6 · Claude 3.5 Haiku · Claude Opus 4.6
- Benchmarks: SWE-bench Verified
Frequently Asked Questions
What is Claude Haiku 4.5?
Claude Haiku 4.5 is Anthropic's fastest, most affordable model, released October 15, 2025. It scores 73.3% on SWE-bench Verified — matching the May 2025 flagship Sonnet 4 — at $1/$5 per million tokens.How much does Claude Haiku 4.5 cost?
$1 per million input tokens and $5 per million output tokens.Specs sourced from Anthropic's Claude Haiku 4.5 announcement. Last updated 2026-06-19.
Claude Haiku 4.5
Claude Haiku 4.5 is a large language model developed by Anthropic.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.