Just trained Tev1 0.8B, a tiny Jev-like classifier. Here it is running completely locally on my mac with @ollama & classifying some tasks. It's extremely fast: only ~50ms E2E latency. Video is not sped up! Releasing weights & benchmarks very soon so you can try it yourself :)

Introducing Figure out which open model is best for your use case. Compare models across coding, agents, long context, vision, finance, and more. Then see how they compare on cost + quality, including what you could save by moving to open models.

I used Jev to classify 1,018 AI research papers. The result: $0.08 total cost and 256ms median end-to-end latency per paper. The pipeline was: 1. Summarize each paper with DeepSeek V4 Flash 2. Send the title + summary + 24 possible topics to Jev 3. Use Jev to classify each paper 4. Visualize everything on The summaries cost $3.99 on @togethercompute. The classifications cost $0.08 on @typesafeai. So for just over $4 of inference, I ended up with a pretty useful way to explore the top AI research papers from the past year. I think this is where things are heading: different models for different parts of the workflow, instead of using one model for everything. I’m running evals on the Jev classifications before replacing the current ones, but the site is already live:

Going to be speaking at AIE New York... with my dad! He's a CPA + business professor which is perfect since the conference theme is finance in AI. Hopefully the first father and son talk at @aiDotEngineer? 😂

This model is insane at landing pages. I asked DeepSeek V4.1 Flash & Claude Fable 5 to build me a landing page for a movie theater. Fable cost $1.21 while V4.1 Flash cost 2.6 cents, making it more than 40x cheaper at similar quality. Gave both the exact same prompt!

Wrote an article on the Open Source AI Stack! It breaks down the ecosystem around open models: harnesses, gateways, routers, tools, inference, & more. A good place to start if you're exploring open source models or just getting into them.

I built a design MCP server for Claude Code, Codex, and OpenCode. It searches hundreds of beautiful websites and gives your coding agent design inspiration. Launching next week.

6 months ago, I started building an internal customer insights tool at Together AI. It pulls in our sales calls, slack channels, & support tickets and turns them into searchable customer insights. Today, I'm open sourcing it. Introducing Open Customer Insights. We use it internally to: ◆ search across customer conversations ◆ surface daily structured insights ◆ track top feature requests in one place ◆ generate monthly reports on specific topics It's become a tool that several people at Together use to keep up with and stay close to our customers. We use it through the web UI or the MCP server. The tech stack is: ◆ @bunjavascript for packages, scripts, & tests ◆ @aisdk for managing model calls & tool calling ◆ @nextjs for the web app and dashboard ◆ @convex for the database, vector search, & crons ◆ @clerk for user auth & MCP auth ◆ @togethercompute for open models & embeddings It also integrates with @Gong_io for calls, @usepylon for support tickets, @ExaAILabs for company enrichment, & @SlackHQ for channels. The code is available here: I hope it's useful to other teams that want to aggregate their customer data & get insights from it! Customize it however you'd like for your team.

Working on a new skill where Fable acts as an orchestrator and delegates work to @opencode subagents running OSS models. It delegates to GLM 5.3 Flash for easier tasks and Kimi K3 for harder tasks, all running on @togethercompute. GLM 5.3 Flash is 100x cheaper than Fable for output tokens while being pretty good at coding so I'm excited to experiment with how well this works.

Introducing I took the top 1k research papers of the last year, summarized them, and visualized them. Fun fact: all 1,000 papers cost a total of $4 to summarize with DeepSeek V4 Flash.

This model is insane at 3D website design. I asked GLM 5.3 (left) & Fable 5 (right) to make me a 3d biking website. GLM nailed it while Fable failed. GLM cost $0.14 while Fable cost $2.21. More than 15x cheaper while being even better in this case!

Fun weekend project: Inference Cards! Trading cards (yugio-inspired) for AI inference & open models. Thinking of printing a physical booster pack to give away as @togethercompute swag, thoughts?

DeepSeek V4 Flash is 6x cheaper than GPT 5.6 Luna for coding tasks! Luna scores higher on DeepSWE, but running Flash twice ($0.20/task) beats Luna once ($0.61/task), while still costing 3x less. Great deepdive on this, worth checking out.

AI Commits just got a lot faster with v3! ◆ 95 file PR went from taking 71s -> 13s, over 5x faster ◆ Full support for local models (@lmstudio & @ollama) ◆ Added support for DeepSeek V4 flash Give it to your agent to get super fast commit messages →

Deepseek V4 Flash 0731 now has the best model intelligence vs cost out of any model. It’s around the same intelligence as GLM 5.2 and GPT Luna while being way cheaper (only $0.14/$0.28 per 1M tokens). Absolutely insane release.