Google DeepMind rolls out Gemini Omni 1.1 Flash

#1 𝕏

Google DeepMind is rolling out Gemini Omni 1.1 Flash to make generative video more controllable, faster to iterate on, and more polished for production-grade use. It can be tried in Flow by Google.

Also covered by: @Google AI

#2 𝕏

Anthropic shared research in which Claude was given 48 hours and 1 GPU to improve the alignment of small models, autonomously researching and proposing methods before training and testing them. Anthropic said the approach worked “surprisingly well,” though specific methods, metrics, and quantitative results were not provided.

Also covered by: @Anthropic

#3 📝 OpenAI News

Our decision on Cursor following its acquisition by SpaceX - OpenAI explains its decision regarding Cursor after Cursor was acquired by SpaceX, outlining the rationale and next steps. The post describes the implications of the acquisition and how OpenAI will proceed.

#4 𝕏

OpenCode announced that Qwen3.8-Flash is now available in OpenCode Go, describing it as multimodal with “125B/6B” and a 1M context.

#5 𝕏

Alexandr Wang said Muse image is live on the Meta Model API at $0.01 per image, describing it as having one of the best price-to-quality ratios for production volumes.

#6 𝕏

Santiago announced that GLM-5.3 is open-weight and can be downloaded from HuggingFace or run using Atomic Agent. He added that the future is open weights.

#7 𝕏

Guillermo Rauch commented on Vercel’s announcement of eve agent creation, contrasting easy agent-building options with approaches that provide a Git repository and ownership of runtime, model choice, skills, tools, connectivity, and sandbox. He described Vercel’s linked page as the easiest way to start with eve.

#8 𝕏

LlamaIndex 🦙 shared experiments to improve static embeddings for retrieval, which offer high throughput but lose accuracy versus traditional models. Raw MaxSim scoring, a small adapter model, and changes to the distillation target and teacher did not produce the desired results.

#9 𝕏

Madhu Guru advised enterprise AI leaders to make their stacks model agnostic by building an evaluation suite today and developing the capability to post-train open models within the next year. These investments enable teams to switch, customize, and compare models on their own workloads while optimizing quality, cost, and latency.

#10 𝕏

Dharmesh Shah said he had soft-launched YouSpot, an AI-native Solo CRM, the previous day via a tweet. The free private beta had a waitlist, while the first 1,000 users could skip the line for $1/month; 547 had already joined.

Also covered by: @David Elkington

#11 𝕏

NVIDIA AI shared a five-minute breakdown of how NVIDIA Dynamo sits around inference engines such as SGLang, vLLM, and TensorRT-LLM to scale inference across GPUs and nodes. The full video is available in the comments.

#12 𝕏

Aravind Srinivas shared that Decagon powers customer support for Delta Airlines, Ticketmaster, Deutsche Telekom, and American Airlines, with Perplexity’s search providing online information for support-related questions.

#13 𝕏

Andrew Ng shared an AI Engineering Skills map focused on software engineering fundamentals in the context of agentic coding.

Also covered by: @DeepLearning.AI

#14 𝕏

NVIDIA AI announced that NVIDIA Warp reached 10 million downloads. Warp enables GPU performance from Python for physics and simulation and is used across computational engineering, geometry processing, and robotics.

#15 𝕏

Santiago recapped a Sapient Intelligence post about the PRAXIST Beta launch, reporting 49 gold medals across 75 tasks versus 35 for Claude Code using Opus 4.8, at ~$3K in token costs versus ~$38K—roughly one-twelfth the cost. Users define a measurable goal and evaluator, while multiple agents explore, test and refine approaches before providing a solution with supporting evidence.

#16 𝕏

Lenny Rachitsky shared templates for three of his favorite @Bot use cases: Be Happier, Talent matchmaker, and Lennybot. Each template links to x.ai.

#17 𝕏

Logan Kilpatrick commented that the developer program page is “a bit clunky” and said he sent feedback. He added that work is underway on automatic redemption without leaving AI Studio; currently, users must already be Google developer program members.

#18 𝕏

Guillermo Rauch said MCP is growing explosively, pointing to growth in Vercel MCP tool calls and mcp-handler npm downloads. He noted that mcp-handler can be used to implement MCP servers.

#19 ▶️

The most expensive software bug in history...

Fireship

#20 𝕏

Peter Yang characterized Claude Cowork and ChatGPT Work as partial solutions, arguing that Grok Bot is the right end state for capable AI agents for non-technical users because it is easily understood as a computer running in the cloud. He suggested most people cannot explain how the other products differ or work.