February 6th news, this morningOpenAI and Anthropic Then we'll publish heavy poundsModelUpdate, GPT – 5.3 - Codex and Claude Opus 4.6。
OpenAI claims that GPT-5.3-Codex was its first model to play a key role in its own research and development, and that the team used an early version to participate in commissioning training, deployment management and evaluation analysis。
The model was updated in a number of assessments, such as SWE-Bench Pro, Terminal-Bench 2.0 and OSWorld-Verified, with a score of 773%, a significant increase from GPT-5.2; and in OSWorld-Verified, the accuracy rate increased to 64.7%, close to the human average。
OpenAI emphasizes that GPT – 5.3-Codex can not only write codes, but also perform complex operations across software, support long-term missions, interact in real time and demonstrate greater autonomy and understanding of intent in the construction of web sites, games, etc。
At almost the same time, Anthropic released Claude Opus 4.6, leading to reasoning, reliability and context processing. It introduced 1M Token context window (Beta) for the first time, with a recall rate of 76% in the MRCR v2 long text search test, well above the previous model。
Opus 4.6 is about 144 Elo higher than GPT – 5.2 in GDP – AA (Assessment of knowledge of high economic value) and is as leading as in Humanity's Last Exam and Brownsecomp。
Anthropic also launched the Agent Teams feature, which allows multiple intelligents to work in parallel to support mission dismantling, independent context and inter-intelligence communication. In the official presentation, 16 Opus 4.6 intelligents independently completed a 100,000-line C-language compiler within two weeks and successfully compiled Linux 6.9 kernels。
In the productivity scene, Anthropic integrates Claude deep into Excel and PowerPoint, and can automatically generate presentations that are consistent with layouts based on tables, and perform multitasking collaborations in Claude Work。
TechCrunch points out that August 4.6's Agent Teams allows ordinary developers to experience the way in which they direct the AI team。
OpenAI emphasizes the high reliability and low variance characteristics of GPT – 5.3 – Codex, which are better suited to the execution and operation of the project, while Anthropic emphasizes the high-ceiling ability and long text-processing advantages of Opus 4.6, which are better suited to the financial, legal and complex decision-making landscape。
Bloomberg reports that the release of Opus 4.6 has hit the financial data services sector, with short-term declines in stock prices for many listed companies。
It is worth noting that according to TechCrunch, OpenAI's GPT-5.3-Codex was planned to be released at the same time as Anthropic, but after 15 minutes of advance release, OpenAI adjusted the release time and quickly updated it online。