Hacker News

Top stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Linux support is coming to Snapdragon X2 Series(qualcomm.com)
    96comments
  2. Claude discovers a novel enzyme system with CRISPR-like repeats(anthropic.com)
    569comments
  3. Feds Target AI Critics as "Foreign Agents"(kenklippenstein.com)
    52comments
  4. Meta VR Glasses(meta.com)
    245comments
  5. ArXiv receives multiyear commitments to support it as an independent nonprofit(arxiv.org)
    11comments
  6. VSCode's SSH Agent Is Bananas (2025)(fly.io)
    93comments
  7. Virtio-nvgpu: Near-native Nvidia GPU access inside a KVM guest(github.com/nestrilabs)
    14comments
  8. The "Windows XP Box" (2003)(mini-itx.com)
    12comments
  9. Fixing the Portobello Police Station Clock(pointinthecloud.com)
    91comments
  10. Mercury 2.5 LLM hits 770 tokens per second(artificialanalysis.ai)
    31comments
  11. We just shipped support for the ugliest part of HTTP: Vary(cloudflare.com)
    17comments
  12. FLAWED's Flaws and What This Means for Industry Research(suhacker.ai)
    1comments
  13. Italian parliament votes for return to nuclear energy(apnews.com)
    399comments
  14. The mystery animal on an ancient god's head(signoregalilei.com)
    19comments
  15. Show HN: Combinators in Array Languages(softwarewrighter.com)
    discuss
  16. Australia says OpenAI agent hacked into government website(channelnewsasia.com)
    18comments
  17. Making Tailscale Faster(tailscale.com)
    40comments
  18. A brief history of Windows scroll bar shortcuts(devblogs.microsoft.com/oldnewthing)
    54comments
  19. LensVLM: Compressing long context as images, expanding only relevant pages(huggingface.co)
    7comments
  20. Show HN: An open-source manufacturing ERP/MES/QMS(carbon.ms)
    8comments
  21. The Curious Power of Punctuation(newyorker.com)
    6comments
  22. Tokens too cheap to meter(jyn.dev)
    185comments
  23. Solving for faster SHA-1 collision detection(sam.dev)
    discuss
  24. Gemini 3.8 text-to-speech(blog.google)
    123comments
  25. Augustofaces: Pareidolia Fine Art(augusto.at)
    2comments
  26. Radicle: Disclosure of Vulnerability in the Network Protocol(radicle.dev)
    50comments
  27. Swap, ZRAM, Zswap and Hibernate on NixOS(matthewbrunelle.com)
    13comments
  28. I don't want the details(michaelheap.com)
    200comments
  29. Z80 REPL (2018)(abagames.github.io)
    18comments
  30. A refined phylochronology of the second plague pandemic in Western Eurasia(pnas.org)
    discuss

Mercury 2.5 LLM hits 770 tokens per second

62 pointsby 5h agoartificialanalysis.ai
31 comments
4h agoHN ↗

The speed means absolutely nothing when it is finishing almost dead last when compared to the frontier AI companies.

4h agoHN ↗

Ah, the old "good, fast, or cheap; pick two" proves true once again.

3h agoHN ↗

Not if your use case needs speed. For one of my products I can't use an LLM that has a p99 of >700ms for TTFT.

2h agoHN ↗

If it could output 1k tokens per second but needed 4 seconds to produce the first batch of 4k, would that not be viable?

2h agoHN ↗

It means something, because it an iterative workflow. If you're willing to burn tokens, it's possible for weaker models to implement tasks by incrementally improving drafts.

4h agoHN ↗

Pricing at $0.25 and $0.75 already puts its cost well above reasonably reputable inference providers for deepseek v4 flash or qwen 3.8-flash-next or similar class of open weight LLMs that fit in under 170GB of RAM, so I don't see the point. I think this is probably also stupider than laguna s 2.1 which can also be very cheap to serve.

1h agoHN ↗

If the model provides me with bad results because it's dumb, I don't care how quickly it does it.

35m agoHN ↗

Is it possible to construct a control system where bad, fast and cheap can become good, fast, and cheap through repeated sampling and a strong spec/eval harness?

I am trying to keep an open mind with AI, but I also have little understanding of control theory, trying to learn.

3h agoHN ↗

If you care about speed Cerebras gpt-oss-120b is 1400tk/s and "just as smart" in ranking.

I've used it on a few for fun projects and its decent but the speed is crazy to watch.

2h agoHN ↗

yeah. K2.6 can run on insane speeds. So sad that they don't have K3 yet.

But it can apparently also run 5.6 Sol

1h agoHN ↗

Yeah but at “call to discuss pricing” rates

2h agoHN ↗

Unfortunately, the lack of an input cache discount makes it prohibitively expensive for most use cases that aren't one-shot prompts.

1h agoHN ↗

Please do not try to use gpt-oss-120b over Cerebras. It is broken, screws up tool calls most of the time, forgets to end thinking blocks and has all sorts of other issues. The speed is amazing but it is absolutely not worth it, especially at that quite incredible cost. Think: $5–10/minute levels of cost with a single agent, because Cerebras also offers no cache pricing for input tokens at all.

1h agoHN ↗

Which is wild because it does, in fact, do caching

9m agoHN ↗

Yea i had some pretty meh results using gpt-oss-120b it in my evals where it should have benefited speed alot but it really under performed what i was expecting.

3h agoHN ↗

I honestly think the diffusion LLM approach is a dead end

It's telling that frontier labs like Google toyed around with it but didn't invest further even for their most speed and cost sensitive small models

Still unclear for what, if any use cases this is pareto frontier

3h agoHN ↗

You can't think that a small startup versus Anthropic's training setup is anywhere near the same scale to make apples to apples comparisons.

Not sure how the Chinese labs pull it off though using autoregressive models. The secret sauce is probably going to be in the training data.

The main reason Google hasn't switched over to DiffusionGemma is because serving at larger batch sizes loses the speed gains you get from diffusion, and most of the primary use case is serving many users at once off a single device with a large batch size.

If you were to move to on-device low latency... like say in a robot or something, then the story might be different...

2h agoHN ↗

Personally, I think it's more that text diffusion is not the ideal driver of an agentic work loop than that text diffusion is a total dead end. I am still hoping to see how it does on authoring and editing with further scaling and optimization. I think the push for AGI has put a bit too much focus on the idea of one general model doing everything.

1h agoHN ↗

DiffusionGemma was released alongside the other gemma-4 models just a few months ago, so clearly google hasn't abandoned the idea.

K2-Horizon-7B has a diffusion and non-diffusion variant, and they claim the same level of intelligence from both models.

42m agoHN ↗

From what I’ve heard, the issue is more that it’s harder to efficiently share the hardware across diffusion requests, so it’s more expensive to serve.

It sounds like there might be opportunities for local models (not open weight, but actually locally run) to use diffusion for faster responses on weaker hardware that doesn’t need to be shared.

But yea, it’s still a red-ish flag that big labs haven’t invested much in it. I could see Google/Apple getting value of this sort of local model, but maybe there’s enough research behind traditional models that it’s not worth the distraction at this point in time.

2h agoHN ↗

this feels like "we got the same benches as gpt-oss-120b but are also potentially slower while saying it is great"

1h agoHN ↗

Mercury 2.5 is below average in intelligence, but well priced when comparing to other models of similar price.

Well priced when compared to other models of similar price, eh?

Are we allowed to call this slop, even if the output is not directly from an LLM?

18m agoHN ↗

At some point the bottleneck becomes tool calling.. and as such, it's preferably if the model is co-hosted (in the same datacenter, at least) with your code repository and all other reference/context it needs (full documentation for most ecosystems, maybe even a copy of common crawl to minimize web fetch usage, etc)

12m agoHN ↗

I used this a few days ago and thought something must be wrong with how fast it was responding. "Mercury 2.5 is below average in intelligence, but well priced when comparing to other models of similar price." this is so funny. So when you have a stupid model that is fast - what do you use it for?