Anthropic's Claude Opus 4.7, announced on April 17, enhances cybersecurity with safeguards against high-risk requests. It improves coding and image handling performance but is less capable than the limited-release Mythos Preview. Feedback from this release will guide future models.

Musk said on the X platform that xAI’s Grok model is catching up to Anthropic’s Claude Opus 4.6, with a prediction it will come close to parity in May and may surpass it in June. He said that this timeline counts as a long time in the AI field. Claude Opus 4.6 is Anthropic’s flagship AI model and is the strongest version in the Claude 4 series.

Musk confirmed on the X platform that Grok 4.2 has 0.5 trillion parameters. Calculations suggest that Claude Opus has about 5 trillion parameters, and Sonnet has 1 trillion. Although Anthropic has not disclosed the model parameters, these figures come from Musk’s remarks. Meanwhile, SpaceX AI’s Colossus 2 supercomputer is training a model with the largest scale of 10 trillion parameters, which will become a key competitive factor.

Gate News message: On April 10, according to Gate market data, as of the time of publication, VVV (Venice) is trading at $8.03. In the past 24 hours, it has pumped 19.99%, with a highest of $8.00 and a lowest pullback to $6.64. The total trading volume over the past 24 hours reached $834,000. Current market cap is approximately $364 million. Venice is a privacy-focused artificial intelligence platform that offers services such as generating conversations and image creation, but does not store user data. Users can privately or anonymously access industry-leading AI models including Claude Opus, GPT-5.2, Flux 2, and more, while also enjoying fully private data protection—every AI prompt is kept on the user’s device and is not stored on servers. In addition, Venice aims to eliminate platform

Zhipu released GLM-5.1, making it available to all GLM Coding Plan users. In the coding ability test, GLM-5.1 scored 45.3, significantly better than the previous generation’s 35.4, and close to Claude Opus 4.6, showing clear progress.

Recently, Anthropic's Claude service experienced a major outage, affecting all three main product lines. Reports show that around 25 incidents occurred between March 13 and 27, with the highest error rate frequently tied to issues with the login system. Although the normal uptime rate remained around 99%, demand for the service and capacity pressure increased noticeably.

The open-source vector database Chroma has released Context-1, a 20 billion parameter search model for multi-turn retrieval. This model utilizes self-editing context technology, trained on multiple tasks through reinforcement learning and a curriculum mechanism, demonstrating outstanding performance, particularly in the fields of web, finance, and law, and also showcasing cross-domain search capabilities in the email sector.

Anthropic is testing a new model called "Claude Mythos," which shows significant improvements in areas such as programming, reasoning, and cybersecurity. Leaked drafts indicate that Mythos is a completely new model tier that surpasses the existing Opus series and possesses leading cybersecurity capabilities. The leak originated from a misconfiguration in the content management system, which led to the exposure of unpublished assets. Anthropic stated that it will adopt a more cautious release strategy.

SWE-rebench is a real-time software engineering task benchmark. The latest leaderboard update removed demo and operation restrictions, and added auxiliary evaluation tasks. Claude Opus 4.6 ranks first with 65.3%, GLM-5 ranks third with 62.8%, and four Chinese models made the top ten, demonstrating progress in Chinese models.

On March 23, SWE-rebench published its latest benchmark rankings, with Claude Opus 4.6 topping the list. This update removed example demonstrations and step limitations, while adding auxiliary evaluation tasks. GLM-5 ranked third among open-source models, with Chinese models collectively occupying four positions.