Compare

Claude Sonnet 4.6 vs Claude Haiku 4.5

Anthropic vs Anthropic. Specs, benchmarks, and real per-task cost — all in one page.

Verdict

Claude Sonnet 4.6 leads on LMSYS ELO (1368 vs 1305). Claude Haiku 4.5 is ~3.8× cheaper on a 3:1 input:output blend.

Anthropic

The workhorse. Near-Opus quality at 1/5 the cost. The default choice for production code and agent workloads.

Excellent cost/quality ratioFast inferenceGreat for Cursor/Claude Code

Anthropic

Fast, cheap, and smart enough for most routing and extraction tasks. Great base model for subagents.

Sub-second latency10x cheaper than OpusStrong structured output

Pricing

Input / 1M$3.00$0.80

Output / 1M$15.00$4.00

Context500K200K

Max output64K16K

LicenseProprietaryProprietary

Released2025-11-042025-10-01

Benchmarks

LMSYS ELO1368.01305.0

MMLU Pro88.781.4

HumanEval92.187.0

SWE-Bench64.549.2

MATH85.076.8

GPQA59.4—

MMMU74.1—

IFEval90.587.3

Per-task cost

Summarize a 1-hour meeting transcript$0.052$0.014

Review a 500-line pull request$0.054$0.014

Answer a customer support ticket$0.021$0.0056

Extract structured data from a resume$0.030$0.0080

Debug a stack trace with context$0.048$0.013

Per-call cost using published token counts for each task. Real-world prompts vary.

Add up to four models, tweak tasks live.