Mercury 3: Diffusion at scale

Pelicans aren't bad, a little quirky: https://tools.simonwillison.net/markdown-svg-renderer?url=ht...

Those were via OpenRouter - here's another set I made by calling the Inception API directly: https://tools.simonwillison.net/markdown-svg-renderer?url=ht...

I still don't understand why Google did effectively nothing with their Gemini Diffusion demo from last year.

It was really cool, and really fast, and just went quiet.

They did release a Diffusion Gemma a few months ago but I've not seen much of a splash from that one, either.

This looks noticeably more capable than previous gen. I purposely gave it a bit of a vague prompt to create an html renderer for UML diagrams, it produced impressive work. Then i asked a question, and in the third prompt I asked it to update the html which it did correctly.

Another thing that stood out, and this might just be a novelty thing, is that it picked different and less painful-to-eyes color schemes for each theme. Congrats to the team behind it.

Question (since their team has reps here and I have never used Inception via api): when you ask for a code change in your codebase (via coding agent), how much code does it need to read? How well does it handle tool calls? Since diffusion is a different way of doing things, I am curious about the code-needed for each incremental change in the codebase, say over a larger task.

I'm very curious what the actual model size is and how the training process differed from a traditional LLM. Cool work regardless, incredible how catching up to a few months behind the cutting edge now seems like an achievable feat for new entrants, if still quite costly.

@volodia (Inception co-founder) has a comment here which is flagged/dead. That's weird; it shouldn't be. Can someone fix that?

love these guys. instead of spending time on something akin to those creepy "dots" or obnoxious @Claude in Slack, they ship actual improvements to speed. you can't imagine what kinds of UX a fast LLM can enable! and for coding I have always been a proponent of strong compile-time type systems and handing off only the actual boilerplate typing, all to maximize actual thinking and speed of high-precision prototyping.

I actually built a small prototype that integrates with JetBrains IDEs - it tracks your in-IDE activity (kinda like JetBrains' own "file-scale tab-completion" feature) but you can invoke Codex with all that context at any moment to let back-then GPT-5.4-nano finish the miniature task. say, you wrote a new use case and your goal is to expose it through a new endpoint, but it's kinda boilerplate-y and DRY is not your style either. then you open a needed file, press a shortcut, tell what you want, and you get it. JetBrains Air kinda absorbed that functionality, Codex tokens are hella expensive (don't get me started on API costs), and JetBrains is ready to front that cost through "Junie Lite" (which is currently Qwen 3.6 27B), so I decided to abandon that prototype. but I do miss the combination of speed and control over resulting code.

I may revisit this idea with Mercury 3, or ask JB to integrate it.

Quick review: It's really fast and works really well in my initial tests. I used it to build a little Swift UI app that runs in the background and helps me OCR and translate text[1]. When I hit CMD + Shift + O it selects a portion of the screen (think of the screenshot tool), runs OCR on the portion and throws that at the google translate API and displays the result in a tooltip above / below the portion. The hotkey / key combination, translation provider and target language can be changed in the settings and there is a little tray icon from which I can access those and enable/disable the tool. I made it come up with a plan for this tool and had it implement the thing end to end in about 8 or 9 steps. It originally messed up the coordinate calculation but quickly corrected that when I pointed it out and other than that it one-shotted all the other steps. The code has tests and looks okay for what it is and it works really well so far. All of it took me around than 45 minutes of prompting, testing, correcting and approving changes. It used 26 million tokens (1.3M uncached) at around 160TPS. I used Oh My Pi as a harness.

[1]When I study chinese or japanese the PDFs I have often don't work with the macos preview selection tool and I don't want to go paste whatever I can select into another tool so I find this useful.

If their benchmarks are to be believed they've made some real progress. I hope we see more open weight diffusion LLMs (currently there's diffusion gemma). Also, if diffusion LLMs are starting to come into their own that's going to be somewhat disruptive isn't it? If these are running at 10x throughput (at least) then that's a lot more tokens on the same hardware.

EDIT: just trying this out on a repo and it's actually pretty good so far. Pleasantly surprised.

I'd be surprised if major labs aren't doing any diffusion experiments, and if they are, it would also be surprising if another lab is ahead of them in diffusion given just how much more compute they have at their disposal. My intuition could be way off though.

Damnit, I've been waiting on an open weights model like this one, still no dice :(

Hmm at least in Openrouter, deepseek v4.1 flash via together/baseten/fireworks still beat mercury 3 on latency and intelligence

Ttft is quite slow with mercury

I've been trialing this for the last few hours. It was quite slow at first, but now I'm getting around 300-400tps. Still nowhere near 1000tps, but it's still incredibly fast compared to what I'm used to with other models. It's quite capable as well. I asked it to present strategies for training an AI to play fighting games and it gave quite well-reasoned output. It searched the existing my existing knowledge base for details about the game's internals, including record/playback, and it's own general knowledge about the game is played. It wasn't sycophantic about my opinions either - it challenged me appropriately.

I think they might have a really decent model if they can deliver their advertised speeds at scale at reasonable prices. A tiered approach could work well.

I tried the model in OpenCode now and it seems great, for certain workloads.

For grilling sessions, it is really fast and the quality of the questions are still decent. The suggested options are less versatile or specific. However, answer3ing one batch of questions and the next batch of questions is there within a few seconds.

And when you don't know how to answer a question, it can write a 600 line html prototype for each option in some 2 seconds and help decide.

I created a pelican worked wonderfully, better than old haiku model or even sonnet models, maybe slightly worse than new gen models.

But overall it's a pretty ok model, the issue with speed seems to be that ttft is still high but still not sure how high thinking burden would be, I hope it scales well will be using it a lot if it works well.

This might actually crush most other LLMs if it's price and performance actually scales well. It was as cheap as deepseek for a few tasks I tried.

Anyone know if there is a subscription for it? I might want to test it out properly without burning a few hundred dollars on it.

I have liked diffusion models for quite a while, ever since i used the original Mercury to play AI Roguelite. I only have one question regarding Mercury 3: how tf did they get it to be so cheap?

If these benchmarks hold up it is a no-brainer against Haiku 5.5 and GPT-6 Luna, especially considering that Haiku 5.5 becomes 5x more expensive past 100k context, and this model goes up to 1M for the same price. (Note: the price increases 2x after a period post launch but is still way lower than Haiku is priced for >100k requests)

Are these models harder to reason about given they produce tokens non-linearly? Especially since we rely on non-latent reasoning to know what the model is "thinking".

I wonder given its speed how competitive it would be with the current batch of decision models like Jev et al. Seems to me for decision models you really want maximum throughput.

Having a chart with a linear scale on one axis and a log scale on the other axis is a really neat trick.

The resulting charts have a high probability of looking fantastic :-D

I'm going to start using this instead of Haiku/GPT-Luna for tasks that require speed.

I'm wondering if it'll still be useful once the discount is lifted?

Looks excellent, looking forward to trying this out. Feels like we're reaching a critical point where these become competitive.

Not as fast as GPT 6.1-Sol on my favorite toy math puzzle. It might be one of those things where intelligence consumes less tokens

I really wish they would make these multimodal. Seems like a definite improvement over 2.5 though in my testing

I wish they had a coding plan so your could kick the tires before paying API rates.

its definitely good at admitting when it doesnt understand stuff. which is a rare treat with these models.

One of the few LLM announcements I still get excited about. dLLMs have so much potential. See, for instance, how gemma diffusion can be repurposed as a decision model with output tokens and diffusion steps set to 1.

The guardrails are ridiculous though:

PROMPT:

> Draw an ascii diagram illustrating how to leave a two storey building in the event of fire

ERROR RESPONSE:

> I'm sorry, but I can't help with this because it involves potentially dangerous or harmful content. I'm happy to help with something else.

Come on

100M free tokens? wow.

Any comparision would make sense only if you made them - Mercury and Luna - the same training data and then the same post process, alignment. It's nice it's fast. I support it. Just saying I don't know how to judge that.

Hi HN, this is Volodymyr, Inception Co-Founder & Chief Scientist.

Today we're launching Mercury 3, our most capable production diffusion LLM.

Quality: Outperforms GPT-6 Luna and Claude Haiku 5.5 across a range of benchmarks at matched reasoning effort. Speed: 1,000+ tokens per second on widely available NVIDIA Blackwell GPUs, 4–8x faster than those models. Context: 1M tokens, up from 260K. Price: $0.10 per million input and $0.35 per million output tokens at launch.

The main takeaway is that diffusion scales. Since Mercury 1 we’ve scaled the model more than 40x in total parameters, and intelligence has gone up at every step without giving up speed. Diffusion generates tokens in parallel through iterative denoising rather than one at a time, so as reasoning models think longer per answer, the speed advantage over autoregressive models of the same size grows.

Early-access partners are running Mercury 3 in browser agents, tutoring apps, real-time search, and customer support. It’s OpenAI API compatible, and it’s also available on OpenRouter, Baseten, and Vercel.

Try it on our API with 100M free tokens.

Happy to dig into the architecture, scaling results, evals, or anything else in the comments.