
MiniMax M3.1 Flash is out, and we've open-sourced a DFlash 2 drafter for it under Apache 2.0. It makes generation up to 3.5× faster, and it's lossless. Congrats to @MiniMax_AI on the release! huggingface.co/incoai/MiniMa…
✓Inference, reimagined for the agentic era

MiniMax M3.1 Flash is out, and we've open-sourced a DFlash 2 drafter for it under Apache 2.0. It makes generation up to 3.5× faster, and it's lossless. Congrats to @MiniMax_AI on the release! huggingface.co/incoai/MiniMa…

Infer/Con is today 🧡 10:55am, Room B: our co-founder @zhijianliu is on the "KV Cache as a Distributed Systems Problem" panel at #InferCon, convened by @datadoghq. At the speaker lunch, he's also hosting a lunch table on speculative decoding and inference for agents. Come join us!

#SGLangSummit was one for the books! Our Founder and Chief Scientist, @zhijianliu shared insights into the technology behind Inco's industry-leading inference speeds. Great conversations and connections all around. @sgl_project

Our co-founder @zhijianliu worked with @jmtang42 and team on PEAR, an open 27B model that learned to use a robot from its own tries in sim, with no demos. It runs locally at 221 tok/s on one RTX 5090 with our DFlash 2 drafter, and 52 tok/s on an M5 Max in Splash ⚡

Day 2 of #SGLangSummit is underway! Stop by booth G6 to meet the @inco_ai team, learn about our industry-leading inference speeds, and chat about the future of #AI inference. See you there! 👋 @sgl_project

Our DFlash 2 just got its driver's license 🚗 @NVIDIADRIVE's new Alpamayo 2 Edge uses it to think faster, right inside the car. Huge congrats! Want to drive even faster? Check out FlashDrive, our #CoRL2026 paper 👇


Cooking up something cool for Splash 👀 Join the community at discord.gg/ZVwTWWyb6h and get a sneak peek before release.

Our co-founder @zhijianliu on why the drafter becomes the bottleneck in speculative decoding, and how parallel drafting gets past it ⚡ DFlash 2 builds on this: about one more accepted token per verification pass (+21%), for just 1.3% more cycle latency.

Inco AI is heading to Infer/Con in San Francisco on Nov 16th, 2026, convened by #Datadog! Our co-founder @zhijianliu is speaking at #InferCon, alongside researchers and engineers building AI systems and tackling the challenges of inference at scale. Building with open models or working on inference? Let’s connect and swap notes on latency, cost, and scaling AI systems. Apply at infercon.ai

Inco is going to #SGLangSummit 2026 🧡 Our open-model endpoints run several times faster than most providers on @ArtificialAnlys. Our co-founder @ZhijianLiu will lead a breakout session on #DFlash and what it takes to build inference for the agentic era, at scale. 🎙️ DFlash and Beyond: How Inco Runs the Fastest Open-Model Endpoints 📍 Nov 12–13, SF 🎟️ luma.com/sglang-summit Coming to the summit? Stop by, say hi, and let's talk about your inference stack.

MiniMax M3.1 Flash is also live on our platform, with the full 1M-token context. platform.inco.ai/

⚡️ Along with this version, we've also released our DFlash 2 drafter for Qwen3.8-Flash-Next! huggingface.co/incoai/Qwen3.…

Congrats to @ornith_ on releasing Ornith-1.5-122B! We trained a DFlash 2 drafter for day-0 support. It speeds up its generation by up to 3×, and it's lossless! huggingface.co/incoai/Ornith…

Congrats @reflection_ai on shipping Beam! Glad to see DFlash 2 powering it from day one. It makes the model decode up to 3× faster with zero quality loss.

Splash is open source under Apache-2.0: github.com/incoai/splash You'll need an M3 or newer and macOS 26.4 or later. To upgrade, run brew upgrade incoai/tap/splash and restart the server. Try it and tell us which Mac you're on and how long your first token takes.