We use Cyber ECI data as of October 8, 2026. The October update expanded the index from 56 to 76 models and from 20 to 35 cybersecurity benchmarks, for a total of 324 model–benchmark results. These come from the UK AI Security Institute (AISI), CAISI, the AI cyber-evaluations company Irregular, reports by model developers, and other public sources. Three of the seven CAISI benchmarks come from its September 2026 assessment of GLM-5.3 and are the only cyber results available for that model.

We include the 57 models with at least two cyber benchmark results, as in the default view of our explorer. The Cyber ECI scores are available to download below, and the underlying benchmark results are available as a CSV.

For each group, we connect the models that scored higher than every earlier model in that group at release. Open-weight models include those released under restricted-use or non-commercial licenses. Closed-weight models include limited-access and unreleased models, such as Claude Mythos 5 and an early checkpoint of Claude Mythos Preview. In the text, Mythos Preview refers to its April 7, 2026 checkpoint.

We compare models by their central estimates. While the Cyber ECI scores in our data explorer currently include 90% confidence intervals for each model, we anticipate methodology changes soon that will let us better estimate uncertainty in differences between models.

To estimate the current lag, we identify the release dates of the closed-weight models whose scores most closely bracket those of GLM-5.3, the highest-scoring open-weight model. GPT-5.4, released on March 5, 2026, scores below GLM-5.3, and Mythos Preview, released on April 7, 2026, scores above it. As of our data cutoff, these releases were 7 and 6 months earlier, respectively.

• Few results per model. GLM-5.3’s score comes from three CAISI benchmark results, and Opus 5.5’s from four benchmarks reported by Anthropic. Opus 5.5 and Astra do not share any benchmarks, and their confidence intervals overlap almost entirely.
• Inferred model identities. CAISI did not name the US models it evaluated, so we attributed those results to the models we judge most likely to have been leading at the time, such as GPT-6 Astra and Claude Mythos Preview.
• Scores move with refits. The October refit lowered several of the highest scores: Claude Mythos 5 fell by 3.9 points and GPT-5.6 Sol by 4.7. Cyber ECI values in our earlier publications are therefore not directly comparable to the new values.
• Coverage. We include only models with at least two cyber benchmark results, so some recent open-weight models with a general ECI do not appear.