What is the highest SWE-bench Verified score?

Claude Opus 5 at 96.0%, ahead of Claude Opus 4.8 and Claude Sonnet 5.

How many models publish a SWE-bench Verified score?

10 of 155. The suite is widely quoted and rarely reported: 145 buyable models publish nothing for it, and 58 publish Terminal-Bench 2.1 instead. Both facts are on this page.

Why not estimate a SWE-bench score for the models that skip it?

Because an estimate would be indistinguishable from a measurement once it is in a table. An absent number is a fact about what the provider chose to publish, and it is recorded here as absent.

The Best LLMs on SWE-bench Verified

Which model solves the most SWE-bench Verified issues?

Only 10 of the 155 buyable models publish a SWE-bench Verified number at all. The other 145 are named in full below, with whatever they publish instead.

Method: Published SWE-bench Verified pass@1, descending, exactly as reported. Every figure here comes from the provider's own announcement, because no independent lab publishes this suite across the catalogue. Pricing as of October 2026. 1 row below carries a score measured at a non-default reasoning tier and is marked as such; every affected model is named on the hub. Read the full method.

Rank	Model	SWE-bench Verified	Terminal-Bench	Quality	Reported by	Blended $/1M
#1	Claude Opus 5Anthropic	96.0	86.1	92.0	Unstated / mixed	$10.00
#2	Claude Opus 4.8Anthropic	88.6	84.6	87.0	Unstated / mixed	$10.00
#3	Claude Sonnet 5Anthropic· non-default tier	85.2	80.5	83.3	Unstated / mixed	$4.00
#4	Nex-N2-ProNex AGI	80.8	67.8	75.6	Unstated / mixed	$1.00
#5	MiMo-V2-ProXiaomi	78.0	not published	78.0	Unstated / mixed	$1.50
#6	Mistral Medium 3.5Mistral	77.6	50.6	66.8	Unstated / mixed	$3.00
#7	Muse Glimmer 30BMeta	76.0	51.7	66.3	Vendor-reported	$0.637
#8	Recursive FlintRecursive	74.2	54.9	66.5	Unstated / mixed	$0.00
#9	Nemotron 3 Ultra 550BNVIDIA	71.9	not published	71.9	Unstated / mixed	$1.00
#10	Qwen3 Coder NextAlibaba	70.6	38.2	57.6	Unstated / mixed	$0.290

Frequently asked

Nearby questions, answered elsewhere

Terminal-Bench 2.1

The suite most of the models below publish instead.

Both suites, weighted

The composite, which does not penalise a model for publishing only one of the two.

All rankings

The hub and the method The Best LLMs by Benchmark Score The Best Value LLMs per Dollar The Best LLMs by Context Window The Cheapest LLMs With a 1M-Token Context Window The Best LLMs on Terminal-Bench 2.1 The Best LLMs on Terminal-Bench Hard The Best LLMs on GPQA Diamond The Best LLMs for Tool Use on tau2-bench