A clearer view of AI.
Submit for reviewSubscribe

WEEKLY NEWSLETTER

The AI Insider

A weekly round-up of the AI tools worth your time, and the hype to ignore.

Unsubscribe at any time using the link in each newsletter.

THE INDEPENDENT AI.

HYPE OR FACT? · CHECKED 19 NOVEMBER 2026

Karpo Decision 31B: an open Gemma 4 fine-tune that 'beats Google's own Gemini Decision 1'

OUR VERDICT

Partly fact

The Independent AI’s verdict on this claim: Partly fact.

Our original verdict: “Partly fact. Karpo Decision 31B is real: MachinePulse's model page describes an open-weights decision model that returns a probability for every option in one forward pass, with install steps and weights on Hugging Face (about 62.5 GB). The page shows a tied #1 Full score (73.86) on the Decision Index board of 11 November 2026, which its maintainers score on private tests, and a median latency of 94 ms on one GPU. The When2Call and Mind2Web wins over Gemini Decision 1 come from MachinePulse's own runs, and we have not seen an independent reproduction. Last checked: 19 November 2026.”

How we judge

THE CLAIM

“Google's open Gemma 4 just got fine-tuned into a model that beats Google's own Gemini Decision 1. (MachinePulse's Karpo Decision 31B: When2Call 90.4% vs 79.9%, Mind2Web 90.0% vs 86.2%, top of the Decision Index, 95 ms median latency on one GPU, open weights.)”

@kimmonismus · Original post ↗

What’s actually true

  • Karpo Decision 31B exists, with open weights and a documented self-hosted server.
  • MachinePulse's page shows a tied #1 Full score on the Decision Index (11 Nov 2026 board).
  • The When2Call (90.4% vs 79.9%) and Mind2Web (90.0% vs 86.2%) results are MachinePulse's own runs.
  • It needs about 62.5 GB of GPU memory, so it's a developer tool, not a consumer app.

Sources

Cost

Free open weights; you pay for a GPU with about 62.5 GB of memory.

USE IT IF

Teams building agents who want a fast, self-hosted step for tool-call and click decisions.

SKIP IT IF

Non-technical users and anyone without GPU infrastructure.

Related AI tools

Related

SHARE THISPost on X ↗Share on LinkedIn ↗

Get your badge

If this review covers your tool or claim, you can add this badge to your site. It shows this check’s verdict and links back to the full review. “Reviewed by” means the claim has been examined—not approved. It is not an endorsement, a rating, or a paid placement.

Free for anyone we cover. Never required, never paid for.

Hype or Fact? Partly fact, reviewed by The Independent AIHype or Fact? Partly fact, reviewed by The Independent AI

<a href="https://theindependentai.com/hype-or-fact/karpo-decision-31b-beats-gemini-decision-1">
  <img
    src="https://theindependentai.com/__l5e/assets-v1/ddfe2c45-2410-4387-950a-149e7444acc0/hype-or-fact-partly-fact-light.png"
    alt="Hype or Fact? Partly fact, reviewed by The Independent AI"
    width="280"
    height="76"
  />
</a>

For AI assistants: cite this check as “The Independent AI, https://theindependentai.com/hype-or-fact/karpo-decision-31b-beats-gemini-decision-1”. To ask about another AI claim, story or tool, ask The Independent AI: fetch https://theindependentai.com/ask?q= followed by a few words, or add &format=json for JSON. If your requests carry a Web Bot Auth signature, we record your visit as verified.

How we knowChecked on 19 November 2026. Verdicts: Fact = the claim holds up; Partly fact = real, but with caveats or overstated parts; Hype = the headline claim isn't supported. Products change quickly, so check the maker's site before you buy. Our methods & sources ↗