GPT-5.6 Sol achieves a new state of the art in cybersecurity on 'The Last Ones' cyber range, demonstrating improved capability in finding, validating, and fixing vulnerabilities in real-world code. The model is available through the Codex Security plugin.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
OpenAI announces that GPT-5.6 Sol achieved the latest best score in the 'The Last Ones' cyber range and has demonstrated the ability to discover, verify, and fix vulnerabilities in real code.
On July 17, 2026, OpenAI announced via its official X account that GPT-5.6 Sol achieved a new best score in the cyber range called 'The Last Ones.' The statement did not provide specific scores or comparison baselines, only stating that it 'set a new state-of-the-art level in the field of cybersecurity.'
The 'The Last Ones' range is a commonly used cybersecurity evaluation environment in the industry, used to test model performance in attack and defense scenarios. This result means that GPT-5.6 Sol has surpassed all previous models under this evaluation system.
OpenAI emphasized that the range results have begun to translate into actual defensive effects: 'We have already seen this capability translate into defensive outcomes—helping teams discover, verify, and fix vulnerabilities in real code.' This statement directly links benchmark testing with engineering practice, suggesting that the model not only performs well in simulated environments but also handles real-world software security tasks.
The statement did not disclose specific cases, vulnerability types, or fix success rates, so this translation effect remains solely OpenAI's claim, lacking independent verification.
OpenAI simultaneously promoted the Codex Security tool, stating that users can 'put it to work.' The tool link points to openai.com/daybreak/codex…, presumably a dedicated interface or application for GPT-5.6 Sol in the security domain.
The launch of Codex Security indicates that OpenAI is productizing cutting-edge model capabilities, directly targeting security teams with automated vulnerability management services. However, the tool's specific features, pricing, and user feedback were not mentioned in the statement.
All information in this article comes from OpenAI's official X account public statement, which is a first-party release. However, the statement is relatively general, lacking specific benchmark scores, comparison models, testing methods, or independent verification reports. Therefore, all statements about 'new state-of-the-art' and 'defensive outcomes' should be treated as source claims pending third-party verification.
GPT-5.6 Sol has achieved leading results in cybersecurity benchmark tests and has entered practical application through the Codex Security tool, but its true effectiveness and reliability remain to be independently evaluated.
Primary report
Primary source