Welcome back! In today’s edition of Data Points, you’ll learn more about:
- FLUX.2, a Nano Banana-class open image generator
- OpenAI’s third-party web analytics security breach
- Genesis Mission, a new U.S. government resource sharing program
- Suno’s latest deal with music giant Warner
But first:
Claude Opus 4.5 achieves SOTA coding and agent capabilities
Anthropic launched Claude Opus 4.5, positioning it as the leading model for software engineering, autonomous agents, and computer use. The model leads on seven out of eight programming languages on SWE-bench Multilingual and scored higher than any human candidate on Anthropic’s internal performance engineering exam within a two-hour time limit. The release includes a new effort parameter that lets developers control token usage; at medium effort, Opus 4.5 matches Sonnet 4.5’s performance while using 76 percent fewer output tokens. Opus 4.5 is available now through Claude apps, API, and major cloud platforms at five dollars per million input tokens and 25 dollars per million output tokens. (Anthropic)
DeepSeekMath-V2 claims International Mathematics Olympiad Gold
DeepSeek introduced DeepSeekMath-V2, a language model that verifies the correctness of its own mathematical reasoning step-by-step rather than relying solely on final answer accuracy. The model achieved gold-level scores on IMO 2025 and CMO 2024 competitions and scored 118 out of 120 on Putnam 2024 when using scaled test-time compute. The system trains a verifier to check theorem proofs for comprehensiveness and rigor, then uses that verifier as a reward model to train a proof generator that identifies and fixes issues in its own work before finalizing solutions. The approach addresses limitations in current reinforcement learning methods that reward correct final answers but can’t verify reasoning quality or handle tasks like theorem proving, which require rigorous derivation without numerical answers. The model’s weights are available freely on GitHub. (GitHub)
Black Forest Labs launches FLUX.2 with multi-reference editing
Black Forest Labs’ image generation model FLUX.2 handles up to 10 reference images simultaneously while maintaining character and style consistency, and edits images at resolutions up to 4 megapixels. The model combines a Mistral-3 24-billion-parameter vision-language model with a rectified flow transformer, improving text rendering, prompt adherence, and photorealism over FLUX.1. The company released four variants: FLUX.2 [pro] and [flex] as managed APIs, FLUX.2 [dev] as a 32-billion-parameter open-weight model available on Hugging Face under a non-commercial license, and FLUX.2 [klein] coming soon as an Apache 2.0 licensed model. The FLUX.2 VAE is available now under Apache 2.0 license, with API access through partners including FAL, Replicate, Runware, TogetherAI, Cloudflare, and DeepInfra. (Black Forest Labs)
OpenAI discloses security incident at external analytics provider
A breach at Mixpanel, a web analytics service OpenAI used for its API platform, exposed user profile data for some API customers. The November 9 incident affected names, email addresses, coarse location data, browser information, and organization IDs associated with platform.openai.com accounts. No chat content, API requests, passwords, API keys, payment information, or government IDs were compromised. OpenAI terminated its contract with Mixpanel and is conducting expanded security reviews across its vendor ecosystem. The company recommends users stay alert for phishing attempts using the exposed profile information and enable multi-factor authentication as a precaution. (OpenAI)
White House “Genesis” to lend government data to AI companies
U.S. President Donald Trump signed an executive order directing the Department of Energy and national labs to build a digital platform consolidating federal scientific data for AI analysis. The Genesis Mission solicits tech companies and universities to apply their AI systems to government challenges in engineering, energy, and national security, including optimizing the electric grid. The administration compared the effort to the Apollo program, though it follows billions in cuts to federal research funding and thousands of job losses among government scientists. Funding comes from the tax and spending bill Trump signed in July, with the project using both national lab supercomputers and private sector computing capacity. (Associated Press)
Suno partners with Warner to train AI models on licensed catalog
Suno signed a deal with Warner Music Group to build new AI music generation models trained on WMG’s licensed recordings. The partnership will introduce features letting fans create music using the voices and likenesses of participating WMG artists, with those artists receiving compensation. Suno will require a paid subscription to download generated songs, though its Studio product will retain unlimited downloads. The company said its core music creation experience will remain unchanged while it develops what it calls a new generation of models that will surpass its current v5 system. Warner Music and Suno also settled their copyright lawsuit; similar cases against Suno by Universal and Sony continue. (Suno)
A special offer for our community
DeepLearning.AI recently launched the first-ever subscription plan for our entire course catalog! As a Pro Member, you’ll immediately enjoy access to:
- Over 150 AI courses and specializations from Andrew Ng and industry experts
- Labs and quizzes to test your knowledge
- Projects to share with employers
- Certificates to testify to your new skills
- A community to help you advance at the speed of AI
Enroll now to lock in a year of full access for $25 per month paid upfront, or opt for month-to-month payments at just $30 per month. Both payment options begin with a one week free trial. Explore Pro’s benefits and start building today!
Still want to know more about what matters in AI right now?
Read this week’s issue of The Batch for in-depth analysis of news and research.
This week, Andrew Ng talked about the potential AI bubble, highlighting underinvestment in AI applications, the need for more AI infrastructure for inference, and the risks associated with AI infrastructure for model training.
“I remain bullish about AI investments broadly. But what is the downside scenario — that is, is there a bubble that will pop? One scenario that worries me: If part of the AI stack (perhaps in training infra) suffers from overinvestment and collapses, it could lead to negative market sentiment around AI more broadly and an irrational outflow of interest away from investing in AI, despite the field overall having strong fundamentals.”
Read Andrew’s full letter here.
Other top AI news and research stories we covered in depth:
- Google led arena leaderboards with Gemini 3 Pro and Nano Banana Pro, showcasing best-in-class multimodal reasoning and image generation capabilities.
- Microsoft and Anthropic formed an alliance, making Claude the first leading language model available from all three cloud giants.
- Record labels backed AI-music startup Klay Image, which secured deals with industry giants Sony, Warner, and Universal.
- Researchers introduced Persona Vectors to help model builders identify and edit out sycophancy, hallucinations, and more.