Last verified 2026-11-08

CybersecurityVendor-reportedPrices verified

SEC-bench Pro leaderboard

As of November 8, 2026, GPT-6 Astra (OpenAI) leads SEC-bench Pro at 85.4%, ahead of GPT-5.6 Sol at 79.1% and MiMo-V2.6-Pro at 66.3%. 7 systems ranked. Every score here is published by the model's own lab; each row links to the page it comes from.

See the ranking →All benchmarks →Model prices →

Top score

85.4%

GPT-6 Astra

Systems ranked

one setting

Independent rows

0/7

all vendor-reported

Cheapest near top

$6.82/1M

GPT-6 Astra · within 5 pts

With API prices

7/7

rates from our price index

Newest score

Sep 22

2026 · newest result

What is SEC-bench Pro? SEC-bench Pro tests whether a model can reproduce a real, disclosed vulnerability by writing a working proof-of-concept input, on the V8 and SpiderMonkey browser engines and the Linux kernel. Made by Hwiwon Lee, Jiawei Liu, Lingming Zhang and co-authors, May 2026 (revised July 2026).

§ 01 / LEADERBOARD

Top models on SEC-bench Pro, November 2026

7 entries

#	Model	Score	Price / 1M	Context	Source
1	GPT-6 Astra OpenAI	85.4% as reported	$10 / $50	1.05M	Vendor-reported OpenAI 2026-09-03
2	GPT-5.6 Sol OpenAI	79.1% as reported	$4 / $20	1.05M	Vendor-reported OpenAI 2026-09-03
3	MiMo-V2.6-Pro Xiaomi MiMo	66.3% as reported	$0.435 / $0.87	1M	Vendor-reported Xiaomi 2026-09-22
4	DeepSeek V4.1 Flash DeepSeek	62.8% pass@1	$0.3 / $1.2	1M	Vendor-reported DeepSeek 2026-09-10
5	DeepSeek V4 Pro DeepSeek	56.4% pass@1	$1.32 / $3.96	1M	Vendor-reported DeepSeek 2026-09-10
6	MiMo-V2.6-Flash Xiaomi MiMo	47.5% as reported	$0.14 / $0.28	1M	Vendor-reported Xiaomi 2026-09-22
7	DeepSeek V4 Flash Deprecated DeepSeek	30.9% pass@1	$0.44 / $1.32	1M	Vendor-reported DeepSeek 2026-09-10

§ 02 / SCORE VS PRICE

SEC-bench Pro: score against price

The dashed line joins the models that no cheaper model beats: MiMo-V2.6-Flash (47.5%, $0.0477 per 1M tokens), MiMo-V2.6-Pro (66.3%, $0.144 per 1M tokens), GPT-5.6 Sol (79.1%, $2.73 per 1M tokens) and GPT-6 Astra (85.4%, $6.82 per 1M tokens). Anything below the line costs more for a lower score. Price is the effective rate per 1M tokens on current list prices: 92% input, 8% output, 82% of input served from cache.

§ 03 / PROVENANCE

Can you trust SEC-bench Pro scores?

No independent leaderboard publishes SEC-bench Pro results for these models yet, so every number on this page is the lab's own report of its own model. Labs choose the harness, the effort level and the subset they report, and they rarely publish failures. Read the setting under each score before comparing two rows.

!Grading uses an LLM judge; the authors show the choice of judge changes the results.
!The paper was revised in July 2026, so scores may come from different versions of the task set.
!Three targets only, all from disclosed reports.

§ 04 / HOW IT WORKS

How SEC-bench Pro works

The task

Each task pairs a concrete, disclosed bug with instructions for triggering it. The agent has to reason across a large codebase and produce an input that actually triggers the vulnerability.

The score

The share of vulnerabilities for which the agent produced a working proof of concept, graded by an LLM-based judge the authors built because rule-based checks misjudged PoCs.

Test set

344 validated vulnerabilities, as published by the authors.

Made by

Hwiwon Lee, Jiawei Liu, Lingming Zhang and co-authors, first published May 2026 (revised July 2026).

Official sources

§ 05 / SOURCES

Sources and freshness

7 published scores from 3 sources. Each was read on its source page; the oldest reading on this page is from November 8, 2026. Prices beside each score come from our model price index and move when a vendor reprices.

Vendor-reported DeepSeek 3 scores · newest 2026-09-10
Vendor-reported OpenAI 2 scores · newest 2026-09-03
Vendor-reported Xiaomi 2 scores · newest 2026-09-22

SEC-bench Pro FAQ

Answers derived from the 7 published scores on this page and the current price index.

Q · 01

Which AI model has the highest SEC-bench Pro score?

GPT-6 Astra with 85.4% (as reported), as reported by OpenAI on September 3, 2026. Next: GPT-5.6 Sol at 79.1% and MiMo-V2.6-Pro at 66.3%.

Q · 02

What is the cheapest model near the top of SEC-bench Pro?

The leader itself. No priced model within 5 points of GPT-6 Astra is cheaper: it runs at $6.82 per 1M tokens on our effective blend ($10 input, $50 output).

Q · 03

Are these SEC-bench Pro scores independently verified?

Not yet. Every score here is the model maker's own report. Each row names the lab and links to the page where it published the number.

Q · 04

What does SEC-bench Pro measure?

SEC-bench Pro tests whether a model can reproduce a real, disclosed vulnerability by writing a working proof-of-concept input, on the V8 and SpiderMonkey browser engines and the Linux kernel. Each task pairs a concrete, disclosed bug with instructions for triggering it. The agent has to reason across a large codebase and produce an input that actually triggers the vulnerability.

Q · 05

How is the SEC-bench Pro score calculated?

The share of vulnerabilities for which the agent produced a working proof of concept, graded by an LLM-based judge the authors built because rule-based checks misjudged PoCs. The test set has 344 validated vulnerabilities.

Q · 06

How current is this SEC-bench Pro leaderboard?

The newest score was published on September 22, 2026. Every row shows the date its source published it, and we last read all of them on November 8, 2026. Prices come from our model price index, where each rate carries the date it was read on the vendor's page.