Anthropic's annual revenue run rate is $65 billion as of the end of July. The company is talking to investors, but its IPO filing remains confidential. Bloomberg originally reported the news. Anthropic's second quarter revenue was about $11.5 billion.
Larry Dignan
Results
Cursor launched a new feature called Origin on all paid plans. Repos, pull requests, code browsing and GitHub sync are available in an early beta. SpaceX recently closed its Cursor acquisition.
According to Bloomberg, Stripe will acquire OpenRouter for more than $7 billion. On the surface, the deal is a bit of a head scratcher, but does validate the importance of model routing.
Anthropic will start putting watermarks in content where Claude played a role. If Claude generated a sentence or paragraph there will be a watermark. In some cases, Claude editing a file can have a watermark. The move has raised a bit of a ruckus and you can follow the comments and storylines at Techmeme.
Hugging Face released its open model report and it's a must read. One big headline is how Qwen-based models account for more than 151,000 derivatives.
According to the report, which is a must read:
"Qwen has become one of the largest foundations in the open model ecosystem. Qwen-based models now account for 151,448 derivatives on the Hub, 2.6× Meta’s total footprint and 4.7× the Llama repositories specifically. Google follows with 82,506 derivatives. The third-largest source is Unsloth, a community account publishing quantized and fine-tuning-ready builds, many of which further extend the Qwen ecosystem.
Qwen derivatives have increased at roughly 180–210 new repositories per day throughout the first seven months of 2026, showing that adoption is not driven only by individual launches. Qwen has become part of the default workflow for developers deciding what models to fine-tune and deploy."
Z.ai released GLM-5.3, a model post-trained on GLM-5.2 to get better coding and cybersecurity capabilities.
In a blog post, Z.ai revealed that GLM-5.3 outperformed GLM-5.2 on multiple benchmarks. The company said every advance in GLM-5.3 was due to post training.
"For GLM-5.3, we pushed environment scaling toward tasks that look less like coding exercises and more like real units of expert work. The environments now cover a much broader range of production workflows, with tasks designed around how engineering and research work is actually carried out in practice."
Reuters is reporting that Silver Lake is in talks to acquire Workday in a deal valued at about $43 billion. The talks are ongoing and there's no guarantee that a deal will be done. The report set off a mini-run in SaaS shares. If Workday looks good to private equity firms, there are dozens of other SaaS vendors that could be for sale.
Google released Gemini 3.7 Flash, its flagship workhorse model. Gemini 3.7 Flash is designed for coding, knowledge work and agents.
The company outlined the usual benchmarks, but the biggest takeaway is the frontier models are competing on price now as enterprises freak out over AI budgets.
Note that Google's pricing in the chart below is introductory. But until Jan. 1, 2027, Gemini 3.7 Flash aims to undercut rivals.
OpenAI named Dali Rajic Chief Revenue Officer. Rajic, formerly of Google's Wiz unit, replaces Denise Dresser, who joined the company in December to build out the enterprise business.
In a blog post, OpenAI said Dresser is leaving to pursue other opportunities. Before OpenAI, Dresser was a long-time Salesforce executive. Rajic was President and Chief Operating Officer of Wiz, which is now owned by Google. He had previous stints at Zscaler and AppDynamics.
Writer, which specializes in marketing and revenue AI workflows, released its Palmyra X6 flagship model and upgrades to its Writer Agent harness. A few points stuck out in the announcement.
- Writer is targeting token spending. The company said Writer Agent operates at a 52% lower cost with 48% improvement in speed when using Palmyra X6.
- Palmyra X6 was trained on top of GLM-5.2.
- Writer benchmarked its latest model and harness compared to frontier models and outperformed.
- Palmyra X6 completes tasks in 26 seconds on average, generates 82 tokens per second and can work unattended for up to 8 hours.
MongoDB launched a set of tools that bring automated embeddings to its Atlas platform via Voyage AI. The company said the Atlas Embedding and Reranking API, voyage-code-4, and vector search in Atlas Stream Processing all are designed to improve retrieval accuracy. MongoDB also announced Atlas Managed MCP Server, a fully hosted service that connects Claude Code, Codex, Grok Build, and Devin to MongoDB Atlas.
Oracle said Oracle Exadata Database Service on Exascale Infrastructure on Oracle AI Database@AWS is now generally available. Oracle and AWS said they have also committed to a new long-term collaboration pact to continue investing in Oracle AI Database@AWS.