Hacker News: volodia

How to build a diffusion language model

volodia — Sun, 30 Aug 2026 23:41:32 +0000

Article URL: https://kuleshov-group.github.io/blog/blog/2026/how-to-build-a-diffusion-language-model/

Comments URL: https://news.ycombinator.com/item?id=49503956

Points: 177

# Comments: 20

Next-Edit in Kilo, Powered by Inception Diffusion LLMs

volodia — Mon, 29 Jun 2026 22:02:21 +0000

Article URL: https://blog.kilo.ai/p/announcing-next-edit-in-kilo-powered-by-inception

Comments URL: https://news.ycombinator.com/item?id=48725881

Points: 2

# Comments: 0

Mercury 2 on PinchBench: Diffusion LLM benchmarked on real OpenClaw agent tasks

volodia — Wed, 25 Mar 2026 18:24:52 +0000

Article URL: https://www.inceptionlabs.ai/blog/mercury-2-on-pinchbench

Comments URL: https://news.ycombinator.com/item?id=47521254

Points: 2

# Comments: 0

New comment by volodia in "Mercury 2: Fast reasoning LLM powered by diffusion"

volodia — Thu, 26 Feb 2026 20:23:41 +0000

Thank you for the detailed feedback! I shared this already with the team.

New comment by volodia in "Mercury 2: Fast reasoning LLM powered by diffusion"

volodia — Wed, 25 Feb 2026 03:15:33 +0000

This looks like an inference glitch that we are working on fixing, thank you for flagging.

New comment by volodia in "Mercury 2: Fast reasoning LLM powered by diffusion"

volodia — Wed, 25 Feb 2026 02:48:03 +0000

There are many ways to do it, but the simplest approach is block diffusion: https://m-arriola.com/bd3lms/

There are also more advanced approaches, for example FlexMDM, which essentially predicts length of the "canvas" as it "paints tokens" on it.

New comment by volodia in "Mercury 2: Fast reasoning LLM powered by diffusion"

volodia — Wed, 25 Feb 2026 02:37:32 +0000

Would love to hear about your experience. Send us an email.

New comment by volodia in "Mercury 2: Fast reasoning LLM powered by diffusion"

volodia — Wed, 25 Feb 2026 02:36:29 +0000

Not imminently, but hard to predict where the field will go

New comment by volodia in "Mercury 2: Fast reasoning LLM powered by diffusion"

volodia — Wed, 25 Feb 2026 02:26:48 +0000

There are few: fast agents, deep research, real-time voice, coding. The other thing is that when you have a fast reasoning model, you spend more effort on thinking in the same latency budget, which pushed up quality.

New comment by volodia in "Mercury 2: Fast reasoning LLM powered by diffusion"

volodia — Wed, 25 Feb 2026 02:21:09 +0000

We agree! In fact, there is an emerging class of models aimed at fast agentic iteration (think of Composer, the Flash versions of proprietary and open models). We position Mercury 2 as a strong model in this category.

New comment by volodia in "Mercury 2: Fast reasoning LLM powered by diffusion"

volodia — Wed, 25 Feb 2026 02:17:55 +0000

That is also our view! We see Mercury 2 as enabling very fast iteration for agentic tasks. A single shot at a problem might be less accurate, but because the model has a shorter execution time, it enables users to iterate much more quickly.

New comment by volodia in "Mercury 2: Fast reasoning LLM powered by diffusion"

volodia — Wed, 25 Feb 2026 02:15:21 +0000

You can think of Mercury 2 as roughly in the same intelligence tier as other speed-optimized models (e.g., Haiku 4.5, Grok Fast, GPT-Mini–class systems). The main differentiator is latency — it’s ~5× faster at comparable quality.

We’re not positioning it as competing with the largest models (Opus 4.5, etc.) on hardest-case reasoning. It’s more of a “fast agent” model (like Composer in Cursor, or Haiku 4.5 in some IDEs): strong on common coding and tool-use tasks, and providing very quick iteration loops.

New comment by volodia in "Mercury 2: Fast reasoning LLM powered by diffusion"

volodia — Wed, 25 Feb 2026 02:11:47 +0000

Thanks for trying it and for the thoughtful feedback, really appreciate it. And we’re actively working on improving quality further as we scale the models.

New comment by volodia in "Mercury 2: Fast reasoning LLM powered by diffusion"

volodia — Wed, 25 Feb 2026 02:09:27 +0000

Thank you for your patience. We are working to handle the surge in demand.

New comment by volodia in "Mercury 2: Fast reasoning LLM powered by diffusion"

volodia — Wed, 25 Feb 2026 02:08:56 +0000

Just to clarify one point: Mercury (the original v1, non-reasoning model) is already used in production in mainstream IDEs like Zed: https://zed.dev/blog/edit-prediction-providers

Mercury v1 focused on autocomplete and next-edit prediction. Mercury 2 extends that into reasoning and agent-style workflows, and we have editor integrations available (docs linked from the blog). I’d encourage folks to try the models!

New comment by volodia in "Mercury 2: Fast reasoning LLM powered by diffusion"

volodia — Wed, 25 Feb 2026 02:03:55 +0000

I’d push back a bit on the Pareto point.

On speed/quality, diffusion has actually moved the frontier. At comparable quality levels, Mercury is >5× faster than similar AR models (including the ones referenced on the AA page). So for a fixed quality target, you can get meaningfully higher throughput.

That said, I agree diffusion models today don’t yet match the very largest AR systems (Opus, Gemini Pro, etc.) on absolute intelligence. That’s not surprising: we’re starting from smaller models and gradually scaling up. The roadmap is to scale intelligence while preserving the large inference-time advantage.

New comment by volodia in "Mercury 2: Fast reasoning LLM powered by diffusion"

volodia — Wed, 25 Feb 2026 01:57:54 +0000

Co-founder / Chief Scientist at Inception here. If helpful, I’m happy to answer technical questions about Mercury 2 or diffusion LMs more broadly.

Mercury 2: Best-in-class speed-optimized intelligence at 1,200 tok/SEC

volodia — Wed, 25 Feb 2026 00:14:54 +0000

Article URL: https://twitter.com/ArtificialAnlys/status/2026360491799621744

Comments URL: https://news.ycombinator.com/item?id=47145489

Points: 1

# Comments: 0

New comment by volodia in "Show HN: Tiny Diffusion – A character-level text diffusion model from scratch"

volodia — Fri, 14 Nov 2025 22:18:47 +0000

There is also this one that was released in October: https://github.com/kuleshov/char-mdlm

New comment by volodia in "Block Diffusion: Interpolating between autoregressive and diffusion models"

volodia — Fri, 14 Mar 2025 19:54:30 +0000

the LLaDA paper is a scaled-up version of this paper; they cite it as an anonymous ICLR submission