Gemini 3 Pro Preview

Tool callingStructured outputReasoningImage inputAudio inputVideo inputFile input

Model card Launch post Website Docs

Preview Gemini flagship for complex reasoning, coding, and rich multimodal prompts

Context: 1.0M
Released: 11mo ago

Sources:models.dev

Benchmarks

DeepSWE is the primary coding benchmark where an exact model evaluation exists. Other sourced evaluations appear below; missing scores are not zero.

Not evaluated in DeepSWE 1.1 as of 2026-09-22. Results are only shown for exact model versions in the official dataset. Official leaderboard.

Other published benchmarks

Evaluation	Reported score and source
ARC-AGI-2	31.1% Third-party evaluator · gemini-3-pro-preview External evaluation via Epoch AI Original-source terms apply
FrontierMath-2025-02-28-Private	37.6% Epoch AI · gemini-3-pro-preview Epoch AI evaluation
FrontierMath-Tier-4-2025-07-01-Private	18.8% Epoch AI · gemini-3-pro-preview Epoch AI evaluation
GPQA diamond	92.6% Epoch AI · gemini-3-pro-preview Epoch AI evaluation
OTIS Mock AIME 2024-2025	91.4% Epoch AI · gemini-3-pro-preview Epoch AI evaluation
SWE-Bench verified	72.9% Epoch AI · gemini-3-pro-preview Epoch AI evaluation
Terminal Bench	62.2% Third-party evaluator · gemini-3-pro-preview Agent: CodeBrain-1 External evaluation via Epoch AI Original-source terms apply
Terminal Bench	69.4% https://www.tbench.ai/leaderboard/terminal-bench/2.0 · gemini-3-pro-preview Agent: Ante https://www.tbench.ai/leaderboard/terminal-bench/2.0 Original-source terms apply
Terminal Bench	62.2% https://www.tbench.ai/leaderboard/terminal-bench/2.0 · gemini-3-pro-preview Agent: CodeBrain-1.5 https://www.tbench.ai/leaderboard/terminal-bench/2.0 Original-source terms apply
Terminal Bench	61.1% https://www.tbench.ai/leaderboard/terminal-bench/2.0 · gemini-3-pro-preview Agent: Droid https://www.tbench.ai/leaderboard/terminal-bench/2.0 Original-source terms apply
Terminal Bench	61.8% https://www.tbench.ai/leaderboard/terminal-bench/2.0 · gemini-3-pro-preview Agent: II-Agent https://www.tbench.ai/leaderboard/terminal-bench/2.0 Original-source terms apply
Terminal Bench	56.0% https://www.tbench.ai/leaderboard/terminal-bench/2.0 · gemini-3-pro-preview Agent: Letta Code https://www.tbench.ai/leaderboard/terminal-bench/2.0 Original-source terms apply
Terminal Bench	65.2% https://www.tbench.ai/leaderboard/terminal-bench/2.0 · gemini-3-pro-preview Agent: SageAgent https://www.tbench.ai/leaderboard/terminal-bench/2.0 Original-source terms apply
Terminal Bench	56.9% https://www.tbench.ai/leaderboard/terminal-bench/2.0 · gemini-3-pro-preview Agent: Terminus 2 https://www.tbench.ai/leaderboard/terminal-bench/2.0 Original-source terms apply

Benchmark compilation: Epoch AI, Capabilities & benchmarking (CC BY attribution). Scores are converted to percentages and model names matched to the Vertical Tabs catalog. Individual evaluations are credited to their original sources. External results may carry separate licences and evaluation configurations; do not interpret scores from different harnesses as directly comparable.

Specifications

Developer: Google DeepMind
Model family: Gemini
Type: Language model
Predecessor: —
Successor: —
Released: November 18, 2025
Model ID: google/gemini-3-pro-preview
Openness: Closed
License: —
Input modalities: text, image, video, audio, file
Output modalities: text
Modality: text+image+video+audio+file->text
Max output tokens: 66K
Knowledge cutoff: 2025-01
Hugging Face: —
Source freshness: Metadata synced Oct 8, 2026, 10:17 PM UTC

Additional source evidence

Source model: google/gemini-3-pro-preview
Source family: gemini-pro
Last updated: 2025-11-18
Knowledge cutoff: 2025-01
Model context: 1.0M
Model output limit: 66K
Open weights: No
File attachments: Yes
Reasoning: Yes
Tool calling: Yes
Structured output: Yes
Temperature control: Yes

Published benchmarks

Benchmark	Score	Evaluation details	Evidence
SWE-Bench Pro	43.3 · resolve rate	public	Source

Complete source record (JSON)

{
  "id": "google/gemini-3-pro-preview",
  "name": "Gemini 3 Pro Preview",
  "description": "Preview Gemini flagship for complex reasoning, coding, and rich multimodal prompts",
  "family": "gemini-pro",
  "last_updated": "2025-11-18",
  "knowledge": "2025-01",
  "attachment": true,
  "reasoning": true,
  "tool_call": true,
  "structured_output": true,
  "temperature": true,
  "modalities": {
    "input": [
      "text",
      "image",
      "video",
      "audio",
      "pdf"
    ],
    "output": [
      "text"
    ]
  },
  "limit": {
    "context": 1048576,
    "output": 65536
  },
  "benchmarks": [
    {
      "name": "SWE-Bench Pro",
      "score": 43.3,
      "source": "https://labs.scale.com/leaderboard/swe_bench_pro_public",
      "metric": "resolve rate",
      "dataset": "public"
    }
  ],
  "release_date": "2025-11-18",
  "open_weights": false
}

Infrastructure

Provider availability

All providers

Provider endpoint data will appear after this model's first OpenRouter endpoint sync.

Text provenance

Gemini app/web

Google SynthID text watermarking

Google DeepMind says text generated in the Gemini app and web experience is watermarked with SynthID by adjusting token probabilities.

Official explanation