Open vs. Closed: GLM-5.2 and DeepSeek V4-Pro Now Trail the Frontier by Just 4–7 Months

On July 20, 2026, the UK's AI Security Institute (AISI) published the first public measurement of the gap between open and closed models: in cybersecurity tasks, it has narrowed to 4–7 months, versus 6–10 months for almost all of 2025.

How far behind are open models?

In AISI's cyber evaluations, the open models GLM-5.2 and DeepSeek V4-Pro performed on par with closed models released 4–7 months earlier. It is the first attempt by a regulator to measure the distance not by eye, but against a fixed set of tasks.

• GLM-5.2 (Zhipu, June 2026) matched Anthropic's Claude Opus 4.6 (February 2026) — a lag of about 4 months
• DeepSeek V4-Pro (DeepSeek) reached the level of Claude Opus 4.5 (November 2025)
• The measurement used AISI's fixed set of cybersecurity tasks
• A year earlier, for most of 2025, the gap stood at 6–10 months
• On long, multi-step tasks, the lag is still more noticeable than on narrow cyber tests

"Recent open models GLM-5.2 and

DeepSeek V4-Pro perform on par with closed frontier models released 4–7 months earlier," — from the AI Security Institute's analytical report.

Why this matters for defense

The main shift is less about speed than about price. According to AISI, a single run of the Cyber Range test with a 100M-token budget costs about $85 on Claude Opus 4.5 or 4.6, roughly $46 on GLM-5.2, and just $1.19 on DeepSeek V4-Pro.

A difference of dozens of times means comparable cyber capabilities are becoming available at a fraction of the cost — and they are landing in open models, which usually have fewer safeguards than closed ones.

"Advanced capabilities are reaching less-protected open models faster than before," — statement from the AI

Security Institute.

Kimi K3 and Hassabis's plan

Jack Clark, in an issue of Import AI, backs up the same trend with two more signals. The first is Moonshot AI's announcement of Kimi K3: a 2.8-trillion-parameter model whose weights and research paper the company promises to release in the coming weeks. According to Import AI's description, in a single 48-hour autonomous run K3 designed, optimized, and verified a chip, and also built a compiler called MiniTriton along the way.

The second signal is a regulatory proposal from Demis Hassabis, co-founder of DeepMind. He proposes creating an industry Standards Body modeled on the US financial regulator FINRA: companies would voluntarily submit a new model for review 30 days before release, and over time the voluntary regime would be written into law.

"The

Standards Body would be responsible for developing evaluation protocols and coordinating with the relevant federal agencies," — Demis Hassabis, DeepMind co-founder, in a post on X.

What this means

Open models are no longer "a year behind." A gap of 4–7 months combined with inference costs that are dozens of times lower shifts the balance: frontier capabilities spread to less-protected models faster, which is exactly why proposals like Hassabis's industry regulator are now on the table.

Frequently asked questions

What is the gap between open and closed models?

It is the amount of time by which the best open models lag behind closed ones on specific tasks. According to AISI's July 20, 2026 measurement, in cybersecurity it stands at 4–7 months, versus 6–10 months a year earlier.

How much cheaper are open models?

In AISI's Cyber Range test, a run on DeepSeek V4-Pro cost $1.19, on GLM-5.2 about $46, and on the closed Claude Opus 4.5 and 4.6 around $85 for the same 100M-token budget.

What does Demis Hassabis propose?

He proposes creating a Standards Body modeled on FINRA: models would be voluntarily submitted for review 30 days before release, and these rules would later be turned into law.