Hacker News

New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Cookie converter and Claude session check I use(claudecookie.com ↗)
    discuss
  2. House Defies Automakers and Mandates Am Radio in New Cars(wsj.com ↗)
    2comments
  3. Netanyahu seeks to strip citizenship as threats to NAZA film directors intensify(theguardian.com ↗)
    discuss
  4. Nature Is Our Learning Environment(periodic.com ↗)
    discuss
  5. American, United, Southwest scale back schedules as fuel shock reshapes flying(reuters.com ↗)
    discuss
  6. Ms. Rachel on Gaza Controversy, YouTube Fame, and Her New Album(rollingstone.com ↗)
    discuss
  7. Open-1B: the first model you don't have to trust(gensyn.ai ↗)
    discuss
  8. OpenAI says CEO Altman will attend Trump's state dinner for Xi(reuters.com ↗)
    discuss
  9. OpenCode Mentor: An anti-vibecoding, programming mentor configuration(github.com/davejpeters ↗)
    discuss
  10. Show HN: Hometowns – Map the area codes in your contacts list(benaduggan.github.io ↗)
    discuss
  11. The LLMs Yearn for the Spines(buttondown.com/hillelwayne ↗)
    discuss
  12. Show HN: unSnag – Display tasks you can do and graph task dependencies(unsnag.app ↗)
    discuss
  13. We're not ready for superpersistent AI(natesilver.net ↗)
    discuss
  14. Show HN: Vending Machine Lab – Build business, leave for 2 wks, see what breaks(vending-machine-lab.pages.dev ↗)
    discuss
  15. Reversing AIs tech job displacement(gist.github.com ↗)
    discuss
  16. Show HN: Interakt – open-source self-hosted search and AI chat for your website(github.com/alphasolutionsrepo ↗)
    discuss
  17. Gnome 51, "A Coruña"(gnome.org ↗)
    discuss
  18. Artemis II Launch 35mm High Speed Film (01.apr.2026)(nasa.gov ↗)
    discuss
  19. A 3D Rasterizer for Embedded Devices(github.com/cubecoders ↗)
    discuss
  20. Breaking the 1.58-bit Barrier for Ternary LLMs(arxiv.org ↗)
    discuss
  21. 2026 Small World in Motion Competition – Nikon Small World(nikonsmallworld.com ↗)
    discuss
  22. RustFS 1.0.0 GA: Production-Ready, Open Source, S3-Compatible Object Storage(rustfs.com ↗)
    discuss
  23. Zuck's bot increased our Vercel bill 10x
    discuss
  24. How VoltDB Works(voltdb.com ↗)
    discuss
  25. Ask HN: People unemployed for >2 years, how do you spend your time?
    2comments
  26. Ukrainian USV Attack on Sochi Reveals Rarely Seen Russian Navy Trained Dolphins(hisutton.com ↗)
    discuss
  27. Copyrightability of LLM-Generated Code(fsfe.org ↗)
    discuss
  28. I'm getting sued by a data center(technically.beehiiv.com ↗)
    discuss
  29. Trump's war on EVs derailed America's auto-factory revival(indiatimes.com ↗)
    discuss
  30. Show HN: A simple Mac extension for recent screenshots(github.com/kanishkvashisht ↗)
    1comments

Xiaomi Mimo 2.6 live post-training dashboard

84 pointsby 1h agomimo.xiaomi.com
25 comments
42m agoHN ↗

Hah, it would be great to see more labs pick this up.

41m agoHN ↗

This is pretty neat. What would be a good reason for the other Model providers to not do this?

30m agoHN ↗

Speculating here, but I assume researchers can make a reasonable estimate of the size of closed models based on factors like training time, training speed, and the number of tokens processed.

Also, Anthropic and OpenAI probably want to keep each other on their toes so they don’t end up on the wrong side of another Opus 4.6 / GPT-5.3-Codex situation, where one lab releases a model only for the other to drop a better one hours later.

38m agoHN ↗

I didn't know 2 thirds of the training data would be source code.

32m agoHN ↗

that is the the "data used to improve the model" when signing up for the subscription plans

32m agoHN ↗

this is the rl run, not the pretraining run

34m agoHN ↗

This is crazy, but sadly anthropic/openai will never do this, what has happened to this world, where chinese companies are more open than US or even EU companies

7m agoHN ↗

Neoliberalism, that famously transparent economic ideology

33m agoHN ↗

Why are they doing this? To try head off accusations about distillation?

26m agoHN ↗

Sometimes you're confident about what you're doing and show how you work to the world.

Keeping the garage door open, or at least making the door translucent. It's always cool.

10m agoHN ↗

That China's official policy is now to prefer open models and open model development may be a part of it.

4m agoHN ↗

With that policy in place, labs might be incentivized to be creative in their openness. This being fun/free PR

32m agoHN ↗

When you run benchmarks while training, isn't that the definition of contamination? Asking because I am not sure if this is normal in big labs now.

30m agoHN ↗

They are using it to evaluate checkpoints during the training, they are probably not using the benchmarks for training the models. It's a common practice for big reinforcement learning runs.

25m agoHN ↗

Kinda yes. The benchmarks become part of the validation set, which means the models get slightly overfit to them if they are used as criteria for stopping the training. But a lot less compared to using them in the training data.

I'd guess everybody uses at least some benchmarks as stopping criteria, which is kinda sensible, but it also does induce some benchmaxxing, and explains partly why the newest models always tend to eke out in benchmarks.

https://en.wikipedia.org/wiki/Training,_validation,_and_test...

21m agoHN ↗

Correct. If just stopping criteria, that is less contaminated. The question gets muddier once you also use it to determine hyperparameters during small-scale runs.

22m agoHN ↗

You gotta have something to aim at. And, presumably, the benchmark is not part of the training data, it is the test against which the model is tested at each stage; is behavior moving in the right direction?

32m agoHN ↗

2.6 Pro: >started 2026-09-15 10:32 UTC

For some reason I thought training took much, much longer than what the progress bar suggests.

This is really neat, I'm currently using mimo 2.5 pro, and it's decent (or great given the price). Hopefully their next one is multimodal.

26m agoHN ↗

These are post-training reinforcement learning steps.

24m agoHN ↗

Yes, updated the submission title to say "post-training" to hopefully prevent further confusion

31m agoHN ↗

I been using MiMo-V2.5 to do most of my work as software engineer, on a variety of projects I'm working on, and I been VERY happy with ROI. The model is very powerful! Not perfect – I've run in hallucination loops once or twice, but nothing a stop-then-continue wouldn't solve.

The cost is unbelievably low, and the quality of intelligence I get is equivalent to when I was working mostly with Anthropic models (late last year/early this year). I'm fully invested in MiMo and I'm very happy with it.

-- PS: I also check almost daily to see if other models are capable of doing such great work. And they do – DS4F is powerful and DS41 is impressive, GLM 5.3 Flash gets a job done well, etc. – but when I add cost of M-token in the ROI math, Jeez! MiMo is an order of magnitude better.

27m agoHN ↗

2.5 Pro or the regular 2.5?

I always found that those Mimo models to be really good at tool calling and following instructions

10m agoHN ↗

I've found that mimo v2.5 works for very basic things like a python script to do one thing, but it also is very 'dumb' compared to qwen 3.8-flash-next (I think the benchmark scores for terminal and coding specific benches back this up). And definitely not in the same class as like a GLM5.2 or 5.3. It's fast but makes basic mistakes that only get caught later.

27m agoHN ↗

You'd think they would make it less obvious that they are running their whole operation with Claude