ClaimsDemis Hassabis

“Excited to launch Gemma 4: the best open models in the world for their respective sizes.”

Demis Hassabis, Google DeepMind CEO, on X 2 April 2026 Post by @demishassabis on X, 2026-04-02 (X)

Factual AI capabilities

Gemma 4 models are the best open models in the world for their respective sizes.

Unresolvable The public record could not settle it.

Factual: Statements about the present or the past, checked against the record: did they hold up?

The second judge would not resolve this claim; the verdict shown is the first judge's.

Said for an organisation they lead or work for; it counts on both records.

What would count as true or false

Judge as of 2026-04-02, the launch date. True: independent benchmarks or leaderboards (e.g. LMArena, Artificial Analysis, standard evaluations) at or soon after launch show each Gemma 4 model size leading all open-weight models of similar parameter count on broad overall measures. Mostly true: Gemma 4 leads in most size classes, or on most broad measures, with a close rival or small exceptions. Mixed: it leads in about half of its size classes, or the results are split across benchmarks. Mostly false: it is competitive but led in only one size class or on narrow measures, while other open models (e.g. Qwen, Llama, Mistral, DeepSeek variants) of similar size generally beat it. False: other open models of comparable size clearly beat Gemma 4 across its sizes on broad measures.

Set before the research began.

Evidence (4)

  1. E1 Sub-32B Open Weights Artificial Analysis · 13 April 2026

    “Qwen3.5 35B A3B (Reasoning) leads on the Intelligence Index at 37, followed by Gemma 4 26B A4B (Reasoning) at 31, GLM-4.7-Flash (Reasoning) at 30”
  2. E2 Gemma 4: Google on the American open weights frontier Artificial Analysis · 6 April 2026

    “Gemma 4 31B (Reasoning, 39) trails Qwen3.5 27B (Reasoning, 42) by 3 points, primarily due to weaker agentic performance.”
  3. E3 Gemma 4 Just Dropped: What Local AI Builders Need to Know InsiderLLM · 2 April 2026

    “The 31B dense variant ranks #3 among open models on LMArena (ELO 1452), and the 26B-A4B MoE punches at nearly the same level (#6, ELO 1441)”
  4. E4 Gemma 4 Just Dropped: What Local AI Builders Need to Know InsiderLLM · 2 April 2026

    “For pure text quality at the 8GB tier, Qwen 3.5 9B wins.”

Reasoning

The evidence shows Gemma 4 31B and 26B A4B trailing Qwen3.5 models of similar size on Artificial Analysis and ranking #3 and #6 among open models on LMArena [E1][E2][E3]. No evidence item gives a result for the E2B or E4B sizes ([E4] does not name a Gemma 4 model), so with two of the four size classes unassessed, the evidence cannot settle which verdict band the criteria point to.

Cite this claim

Demis Hassabis, “Gemma 4 models are the best open models in the world for their respective sizes.” (said 2 April 2026). Verdict: unresolvable. Reliability API, claim c_180019ad3c83, method 0.1.0+c876030c, dataset release v1.1.2. https://epistacklabs.com/claims/c_180019ad3c83

This claim is in dataset release v1.1.2, with the same verdict.

Reply to this claim

For Demis Hassabis, or someone speaking for them: a correction or counter-evidence. Replies are read before anything is shown; an approved reply shows your name, role, date, statement and link. Your email address is never shown and is deleted 30 days after the decision (privacy).

More factual claims by Demis Hassabis

  1. Mostly true Factual 20 May 2026
    “Performs better than 3.1 Pro on coding & agentic tasks”

    Gemini 3.5 Flash performs better than Gemini 3.1 Pro on coding and agentic tasks.

  2. Mostly true Factual 19 May 2026
    “It can take photos, video & audio and build entirely new scenes.”

    Gemini Omni can take photos, video and audio and build entirely new scenes.

  3. Mostly true Factual 10 April 2026
    “I didn’t feel like that was proper AI because that system, let’s take Deep Blue, okay, it’s world champion level at chess, at chess, but it can’t do anything else.”

    Deep Blue was world-champion level at chess but could not do anything else.

  4. Resolved true Factual 10 April 2026
    “And it uses AI to actually digitally watermark any generated image.”

    SynthID uses AI to digitally watermark any generated image, and Google generative technologies such as Veo carry this watermarking.

  5. Mostly true Factual 3 August 2025
    “They still can't really yet go beyond asking a new novel question or a new novel conjecture or coming up with a new hypothesis that has not been thought of before.”

    Current AI systems still can't really yet go beyond asking novel questions or conjectures, or come up with new hypotheses not thought of before.

All of them

Verdicts are made automatically; how verdicts are made.