Gemini 3 Pro Model Card [pdf] (storage.googleapis.com)

> The training dataset also includes: publicly available datasets that are readily downloadable; data obtained by crawlers; licensed data obtained via commercial licensing agreements; user data (i.e., data collected from users of Google products and services to train AI models, along with user interactions with the model) in accordance with Google’s relevant terms of service, privacy policy, service-specific policies, and pursuant to user controls, where appropriate; other datasets that Google acquires or generates in the course of its business operations, or directly from its workforce; and AI-generated synthetic data.

Well don't complain when you are using Gmail and your emails are being trained to develop Gemini.

gone now;

wayback machine still has it: https://web.archive.org/web/20251118111103/https://storage.g...

For the veracity of the link itself: https://storage.googleapis.com/deepmind-media/* has been used by DeepMind itself (e.g. "View tech report" in https://deepmind.google/models/gemini/) so it is a genuine leak.

it was accidentally pushed a little early, and now it has been taken down.

here’s the archived pdf: https://web.archive.org/web/20251118111103/https://storage.g...

They scored a 31.1% on ARC AGI 2 which puts them in first place.

Also notable which models they include for comparison: Gemini 2.5 Pro, Claude Sonnet 4.5, and GPT-5.1. That seems like a minor snub against Grok 4 / Grok 4.1.

good benchmark stats except for coding where it looks similar to other SOTA models

Benchmark suggests it is a resounding win for Gemini 3 Pro as the top model.

If these numbers are true then OpenAI is probably done, Anthropic too. Still, it's hard to see an effective monetization method for this tech and it clearly is eating Google's main pie which is search.

It says it's been trained from scratch. I wonder if it will have the same undescribable magic that makes me spend an hour every day with 2.5. I really love the results I can get with 2.5 pro. Google eventually limiting aistudio will be a sad day.

Also I really hoped for a 2M+ context. I'm living on the context edge even with 1M.

So does google actually have a claude console alternative currently?

It's hilarious that the release of Gemini 3 is getting eclipsed by this cloudflare outage.

Benchmarks from page 4 of the model card:

    | Benchmark             | 3 Pro     | 2.5 Pro | Sonnet 4.5 | GPT-5.1   |
    |-----------------------|-----------|---------|------------|-----------|
    | Humanity's Last Exam  | 37.5%     | 21.6%   | 13.7%      | 26.5%     |
    | ARC-AGI-2             | 31.1%     | 4.9%    | 13.6%      | 17.6%     |
    | GPQA Diamond          | 91.9%     | 86.4%   | 83.4%      | 88.1%     |
    | AIME 2025             |           |         |            |           |
    |   (no tools)          | 95.0%     | 88.0%   | 87.0%      | 94.0%     |
    |   (code execution)    | 100%      | -       | 100%       | -         |
    | MathArena Apex        | 23.4%     | 0.5%    | 1.6%       | 1.0%      |
    | MMMU-Pro              | 81.0%     | 68.0%   | 68.0%      | 80.8%     |
    | ScreenSpot-Pro        | 72.7%     | 11.4%   | 36.2%      | 3.5%      |
    | CharXiv Reasoning     | 81.4%     | 69.6%   | 68.5%      | 69.5%     |
    | OmniDocBench 1.5      | 0.115     | 0.145   | 0.145      | 0.147     |
    | Video-MMMU            | 87.6%     | 83.6%   | 77.8%      | 80.4%     |
    | LiveCodeBench Pro     | 2,439     | 1,775   | 1,418      | 2,243     |
    | Terminal-Bench 2.0    | 54.2%     | 32.6%   | 42.8%      | 47.6%     |
    | SWE-Bench Verified    | 76.2%     | 59.6%   | 77.2%      | 76.3%     |
    | t2-bench              | 85.4%     | 54.9%   | 84.7%      | 80.2%     |
    | Vending-Bench 2       | $5,478.16 | $573.64 | $3,838.74  | $1,473.43 |
    | FACTS Benchmark Suite | 70.5%     | 63.4%   | 50.4%      | 50.8%     |
    | SimpleQA Verified     | 72.1%     | 54.5%   | 29.3%      | 34.9%     |
    | MMLU                  | 91.8%     | 89.5%   | 89.1%      | 91.0%     |
    | Global PIQA           | 93.4%     | 91.5%   | 90.1%      | 90.9%     |
    | MRCR v2 (8-needle)    |           |         |            |           |
    |   (128k avg)          | 77.0%     | 58.0%   | 47.1%      | 61.6%     |
    |   (1M pointwise)      | 26.3%     | 16.4%   | n/s        | n/s       |

n/s = not supported

EDIT: formatting, hopefully a bit more mobile friendly

Trying to open this link from Italy leads to a CSAM warning

Curiously, this website seems to be blocked in Spain for whatever reason, and the website's certificate is served by `allot.com/emailAddress=info@allot.com` which obviously fails...

Anyone happen to know why? Is this website by any change sharing information on safe medical abortions or women's rights, something which has gotten websites blocked here before?

There needs to be a sycophancy benchmark in these comparisons. More baseless praise and false agreement = lower score.

Hopefully this model does not generate fake news...

https://www.google.com/search?q=gemini+u.s.+senator+rape+all...

[flagged]

What does the "Google Antigravity" mean? The link is http://antigravity.google/docs, seemingly a new product but now routing to the Google main page.

Archive link: https://web.archive.org/web/20251118111103/https://storage.g...

Title of the document is "[Gemini 3 Pro] External Model Card - November 18, 2025 - v2", in case you needed further confirmation that the model will be released today.

Also interesting to know that Google Antigravity (antigravity.google / https://github.com/Google-Antigravity ?) leaked. I remember seeing this subdomain recently. Probably Gemini 3 related as well.

Org was created on 2025-11-04T19:28:13Z (https://api.github.com/orgs/Google-Antigravity)

Interesting to see on page 2 the reference to ML pathways [1]. Looks like a multi layer mixture of experts. Is this common ?

[1] https://blog.google/technology/ai/introducing-pathways-next-...

I know this is a little controversial but the lack of performance on SWE-bench is hugely disappointing I think economically. These models don’t have any viable path to profitability if they can’t take engineering jobs.

     This model is not a modification or a fine-tune of a prior model

Is that common to mention that? Feels like they built something from scratch

It is interesting that the Gemini 3 beats every other model on these benchmarks, mostly by a wide margin, but not on SWE Bench. Sonnet is still king here and all three look to be basically on the same level. Kind of wild to see them hit such a wall when it comes to agentic coding