Intel / post

Open-weights DeepSeek matches gated GPT 6 Sol on offensive cyber range

Source: x.com
Date: Published Oct 20, 2026

Context

Vals AI reports that DeepSeek's V4.1 Flash 1017 beat OpenAI's GPT 6 Sol on a multi-step offensive cybersecurity task, part of a Vals is building from simulated corporate networks. Vals presents the benchmark as an indicator of how much cyber risk a model would pose in the hands of a bad actor. It says the full will be made public in the coming weeks and that this post shares only its first finding, so the post's text gives no scores, margin or task details.

The comparison is notable for how the two models were reached. Vals says it tested Sol through Daybreak, OpenAI's trusted access program. In its August Daybreak announcement, OpenAI described the program as giving approved defenders access to its models with the system-level safeguards that screen ordinary cybersecurity requests removed, plus a tier for purpose-trained cyber models; Vals does not say which access it used. DeepSeek's V4.1 Flash 1017, by contrast, was published on Hugging Face with its weights available for download. Until the full evaluation appears, the result is Vals' own preliminary finding.

Terms in this piece · Glossary

benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.

Read the source x.com