Claude Haiku 4.5 vs Claude Sonnet 4.5

Benchmarks, API pricing and specs, head to head. Data updated 2026-06-10.

Claude Haiku 4.5

Anthropic · Oct 2025

Anthropic's fast tier: Sonnet-4-class coding performance at $1/$5 and more than twice the speed.

Claude Sonnet 4.5

Anthropic · Sep 2025

65.5

The model that made 30-hour autonomous coding sessions real — 77.2% SWE-bench Verified and a 1M-token context beta at $3/$15.

The verdict

Claude Sonnet 4.5 wins 3 of the 3 benchmarks these models share, against 0 for Claude Haiku 4.5. On price the gap is dramatic: Claude Haiku 4.5 works out roughly 3.0x cheaper per blended million tokens.

Benchmark head-to-head 03

SWE-bench Verified
73.3%77.2%
Claude Haiku 4.5Claude Sonnet 4.5
GPQA Diamond
~73%83.4%
Claude Haiku 4.5Claude Sonnet 4.5
Terminal-Bench
~41%50%
Claude Haiku 4.5Claude Sonnet 4.5

Specs & pricing

Claude Haiku 4.5Claude Sonnet 4.5
modhub Index65.5
Input price / 1M$1$3
Output price / 1M$5$15
Context window200K200K
Max output64K64K
Open weightsnono
Reasoning modelyesyes
Multimodal inputtext, imagetext, image
Knowledge cutoffFeb 2025Jan 2025
ReleasedOct 2025Sep 2025
Example monthly cost*$17.50$52.50

* 10M input + 1.5M output tokens per month at list prices, no caching. Green = better value on that row.

Frequently asked questions

Which is better, Claude Haiku 4.5 or Claude Sonnet 4.5?
Claude Sonnet 4.5 wins 3 of the 3 benchmarks these models share, against 0 for Claude Haiku 4.5. On price the gap is dramatic: Claude Haiku 4.5 works out roughly 3.0x cheaper per blended million tokens.
Which is cheaper, Claude Haiku 4.5 or Claude Sonnet 4.5?
Claude Haiku 4.5 costs $1/$5 per million input/output tokens, while Claude Sonnet 4.5 costs $3/$15. For a typical workload of 10M input and 1.5M output tokens per month, that's $17.50 versus $52.50.
Which model is better for coding, Claude Haiku 4.5 or Claude Sonnet 4.5?
On SWE-bench Verified — the standard agentic-coding benchmark — Claude Sonnet 4.5 scores 77.2% versus 73.3% for Claude Haiku 4.5, making Claude Sonnet 4.5 the stronger pick for coding agents.

More comparisons