Compare

Claude Sonnet 4.6 vs o4

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

Verdict

Claude Sonnet 4.6 is ~4.4× 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

OpenAI

Deep-reasoning model. Spends more tokens thinking to crush math, science, and hard coding problems.

State-of-the-art math and scienceExtremely high-ceiling reasoning

Pricing

Input / 1M$3.00$15.00

Output / 1M$15.00$60.00

Context500K200K

Max output64K100K

LicenseProprietaryProprietary

Released2025-11-042026-02-18

Benchmarks

LMSYS ELO1368.0—

MMLU Pro88.793.8

HumanEval92.192.3

SWE-Bench64.568.0

MATH85.096.2

GPQA59.482.5

MMMU74.1—

IFEval90.5—

Per-task cost

Summarize a 1-hour meeting transcript$0.052$0.255

Review a 500-line pull request$0.054$0.240

Answer a customer support ticket$0.021$0.093

Extract structured data from a resume$0.030$0.135

Debug a stack trace with context$0.048$0.210

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

Add up to four models, tweak tasks live.