Hacker News

New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Why did Google declare war on HTTP? (2016)(josh.com)
    1comments
  2. Show HN: NetHackers(dunnolab.ai)
    discuss
  3. Sidehoe – Your Roster. Handled(sidehoe.chat)
    discuss
  4. That About Wraps It Up for Stock Mac UI(inessential.com)
    discuss
  5. Meta Connect 2026(meta.com)
    discuss
  6. I tried using Jev for judgment based guardrails(github.com/deepansh-saxena)
    discuss
  7. Why Don't My Agents Break Containment?(harper.blog)
    discuss
  8. Show HN: Recipe Card Creator(sunnyshorescreative.com)
    discuss
  9. Show HN: Hosted JayBase – An append-only data store for safe AI agent writes(jaybase.ai)
    discuss
  10. Trump Invites Putin to G20 in Miami, Rubio Says(politico.com)
    discuss
  11. Critical GitLab Auth RCE via Double-Free in Regex Parser(docs.gitlab.com)
    discuss
  12. Ursa: A storage engine that implements Lakestream(github.com/openlakestream)
    discuss
  13. Show HN: Make cursed fonts like Times New Bastard(mitpit.com)
    discuss
  14. I made a social media site
    discuss
  15. Show HN: The Mouse Is Coming Too – Easy Switch for Legacy Logitech Devices(github.com/ensoenso)
    discuss
  16. Show HN: PromptSpend – LLM API prices re-checked daily, with source per price(promptspend.com)
    1comments
  17. Do Our Emotions Have a Culture?(booksandideas.net)
    discuss
  18. ArXiv receives multiyear commitments to support it as an independent nonprofit(arxiv.org)
    discuss
  19. ASML currently sells no chipmaking machines in Europe, executive says(nltimes.nl)
    discuss
  20. AI leaders warn UN of security risks as systems grow more powerful(reuters.com)
    discuss
  21. In-Process MCP for MySQL(villagesql.com)
    1comments
  22. Researcher Found 76 Vulnerabilities in Managed PostgreSQL Security Extensions(mehmetince.net)
    discuss
  23. Linux support is coming to Snapdragon X2 Series(qualcomm.com)
    7comments
  24. Toyota is finally taking its best-seller electric(electrek.co)
    discuss
  25. Getting the most out of Opus 5.5 in Claude and Claude Code / claude.dev(claude.dev)
    discuss
  26. A new world airport and its baggage(computer.rip)
    discuss
  27. Wasmer Swift SDK: Run Node.js, Python and FFmpeg Sandboxes on iOS(wasmer.io)
    discuss
  28. Open Source Opinions(stanceoftheday.com)
    discuss
  29. FBI Hack Exposed FBI's Own Hacking Unit(404media.co)
    discuss
  30. Google's Summer of Love(demandsphere.com)
    discuss

Mercury 2.5 LLM hits 770 tokens per second

4 pointsby 52m agoartificialanalysis.ai
3 comments
44m agoHN ↗

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

5m 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.