I think langgraph/smith is generating solid revenue though last public report was a bit ago
LLMS
-
Introducing arXivQA: Training retrieval agents for arXiv search
By
–
Introducing arXivQA: Training retrieval agents for arXiv search We curate a multi-hop arXiv dataset based on real queries and use rubric-as-rewards to train a Qwen model for production-grade retrieval Our latest blog details our experience training with both RLVR and rubrics
-
Karpathy’s AutoResearch Enables AI Self-Improvement Loop
By
–
You can now give an AI a GPU and let it improve its own model overnight. This repo turns LLM research into an automated search loop. @karpathy just open-sourced a system that lets AI iterate on its own training code. Karpathy’s AutoResearch turns LLM improvement into an
-
OpenAI’s Recent History: ChatGPT, Leadership Changes, and Legal Battles
By
–
3 new reports and an editorial @NatureCancer reinforce the benefit of AI for breast cancer screening https://
nature.com/articles/s4301
8-026-01126-1
… https://
nature.com/articles/s4301
8-026-01127-0
… https://
nature.com/articles/s4301
8-026-01128-z
… https://
nature.com/articles/s4301
8-025-01109-8
… -
Self-Evolving AI Agents Tool Genesis Benchmark Research
By
–
New research on Self-Evolving AI Agents. Really interesting benchmark for evaluating a critical but overlooked capability: can LLMs create reusable tools from scratch, not just use existing ones? Tool-Genesis tests whether models can infer interfaces, generate schemas, and
-
Portfolio Strategy: Specialized Models Beat One-Size-Fits-All
By
–
The winning strategy is not one model to rule them all. It’s a portfolio: • reasoning models
• fast models
• research models
• coding agents
• multimodal systems
A model for every job. -
Structured-RAG: Solving Aggregative Queries in RAG Systems
By
–
Standard RAG falls apart on aggregative queries. Things like "average ARR for companies with >1k employees" require reasoning across hundreds of docs, and vector retrieval just can't handle that reliably. We built Structured-RAG to fix this:
– induces a schema from your -
Run Claude Code with local LLMs for free
By
–
Run Claude Code using local LLMs for FREE. No API costs. No data leaving your machine. Here's how it works: Claude Code lets you swap its backend via a single env variable. Point `ANTHROPIC_BASE_URL` to a local llama.cpp server, and it'll route all requests to whatever model
-
BFCL Score Improvements Enable Cleaner Multi-Step AI Workflows
By
–
Yeah the BFCL score difference is wild. In practice it means fewer broken tool calls and cleaner multi-step workflows
-
MiniMax M2.5 Open Source: Claude Opus Performance at 95% Lower Cost
By
–
MiniMax M2.5 went fully open source. If you're running OpenClaw, this changes the game. It's basically Claude Opus performance but 95% cheaper. It scores 80.2% on SWE-Bench Verified. OpenClaw's been great for persistent AI agents with memory, tools, and messaging integrations.