Rohan Varma details first-week updates and roadmap for dots
One week after the launch of dots, Rohan Varma detailed performance and bug fixes alongside upcoming plans for deeper Codex app integration and local computer control.
AI News
Google and MIT introduce Coco agent platform for hardware and ML model co-design
Researchers from Google and MIT have introduced Coco, an agent platform designed for TPU architects to co-design machine learning models and accelerators using relational database simulation sweeps and typed tools.
SpaceX reportedly seeks $40 billion to buy Nvidia AI chips
According to the Financial Times, SpaceX is reportedly seeking $40 billion to fund purchases of Nvidia AI chips, with Apollo leading the financing.
Polymarket puts the chance of OpenAI announcing another Millennium Prize solution by end of year at 24%
Polymarket estimates a 24% chance that OpenAI will announce another Millennium Prize solution by the end of the year.
OpenAI releases mathematical results from internal frontier model on GitHub
OpenAI has officially published a broad range of new mathematical results generated by an internal frontier model. The release follows consultations and public recommendations from the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study.
Hao AI Lab brings FastVideo FastH3 to consumer hardware and releases FastH3 Trim
Hao AI Lab announced that FastVideo FastH3 runs on single consumer machines, including NVIDIA RTX GPUs and Apple Silicon, and released the compact FastH3 Trim model requiring 8 GB of memory.
OpenCode launches Mistral Large 4 with two-week 50% discount
OpenCode has made Mistral Large 4 available on its platform, offering 50% off the model for the next two weeks.
CoreWeave launches Forge to connect agent tracing, evaluation, and serverless RL
CoreWeave has launched CoreWeave Forge, an environment that links agent tracing, evaluation, and serverless reinforcement learning. The platform offers a free starting tier and a 30-day Pro trial on new accounts.
Ai2 releases technical report on supercharging Olmo-core for MoE training
Ai2 has published a technical report titled 'Supercharging Olmo-core for Efficient and Scalable MoE Training,' focusing on scalable and efficient MoE model training.
Google DeepMind's EmbeddingGemma 2 is now available on Ollama
Ollama has added support for Google DeepMind's EmbeddingGemma 2, allowing developers to run the multimodal on-device embedding model locally.
WHOAFRO uses Google Earth AI agent prototype to map Ebola risk in DRC
WHOAFRO deployed Google Earth AI's Geospatial Reasoning agent prototype during the Ebola outbreak in the Democratic Republic of Congo, identifying 48 exposed settlements and over 45,500 at-risk people in minutes.
Zyphra trains ZAYA1-8B reasoning model from scratch on full-stack AMD platform
Zyphra trained its ZAYA1-8B reasoning model from scratch on a full-stack AMD platform, utilizing open software libraries and direct collaboration with AMD to maintain efficient compute scaling.
Every releases Slack agent built on Claude Managed Agents to subscribers
The Every team built a company agent on Claude Managed Agents for Slack to share skills across new models, making it available to subscribers following internal use.
OpenAI opens Decisions API in public beta to all developers
OpenAI has launched the Decisions API in public beta for all developers, stating it allows applications to select models, tools, or actions up to 10 times faster than GPT-6 Luna through the Responses API.
v0 adds ChatGPT subscription support to iOS app
Users can now connect their ChatGPT Plus or Pro subscriptions to build with their ChatGPT tokens inside the v0 iOS app.
Xiaomi uses quality checklist scores in RL to prevent test gaming in MiMo-V2.6-Pro
Xiaomi addressed an issue where models trained only to pass tests added unrequested code and let errors pass silently by multiplying test results by quality checklist scores. The resulting MiMo-V2.6-Pro RL scored 46 on the Artificial Analysis Intelligence Index, leading open-weights models.
Google's Nano Banana 2.1 joins Arena.ai image generation and editing leaderboards
Arena.ai has added Google's Nano Banana 2.1 to its Text-to-Image, Image Edit, and Multi-Image Edit arenas, where the model placed fourth in Multi-Image Edit, fifth in Text-to-Image, and sixth in Image Edit.
Splunk Agent Observability adds cross-tool AI token tracking and billing prediction
Splunk has introduced new capabilities in Splunk Agent Observability that let teams track token usage and costs across AI coding tools in one place and predict final billing before the cycle ends.
Mistral Large 4 outperforms Opus 5.5 and GPT-6 Astra on cybersecurity benchmarks
Cline reported that Mistral Large 4 outperformed Opus 5.5 and GPT-6 Astra on cybersecurity benchmarks, largely because the competing models had approximately 40% of tasks blocked by safety filters.
Cline enables Mistral Large 4 for security workflows in CLI and desktop app
Cline has added Mistral Large 4 support across its command-line interface and desktop application. The team stated the model enables security-focused workflows previously blocked or unavailable when using GPT or Claude models.