Google, released Nov 18, 2025

Gemini 3 Pro Previewprice, context, benchmarks and release details

Provisional: not enough results to rank yet 68% confidence 68 percent, Medium confidence, 5 of 9 expected sources in
58.6
SI Score
#68 of 142 ranked models
Input, per 1M tokens
not yet reported
Output, per 1M tokens
not yet reported
Context window
1Mmodels.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗
Max output
65.5Kmodels.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗
Released
Nov 18, 2025models.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗

How the score breaks down

Coding (weight 40 percent) 63.7
Math (weight 15 percent) 91.4
Preference (weight 15 percent) —
Reasoning (weight 30 percent) 55.9

Weights: reasoning 30%, math 15%, coding 40%, preference 15%. Results use fixed 0–100 scales before averaging, and thin evidence is pulled toward 50. Method si-v3-retained-evidence-2, computed Oct 9, 2026, 06:15 UTC.

Around it on the leaderboard

  1. 66 Kimi K2.5 59.0
  2. 67 Qwen3.7 Max 58.6
  3. 68 Gemini 3 Pro Preview 58.6
  4. 69 GPT-5.5 Instant 58.3
  5. 70 MiniMax-M2.7 57.9

Full leaderboard

Benchmark results

5 benchmarks, 8 results

Each row shows the best published result. Where a model was tested at several settings, such as reasoning effort, open the row to see each one. Hover or tap a value for its source.

GPQA Diamondreasoning 92.6%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] Published Nov 19, 2025 Retrieved Oct 9, 2026 · CC-BY
Open source ↗ 92.6
Humanity's Last Exam (Scale AI)reasoning 37.5%Humanity’s Last ExamPotential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions. [variant] Published Nov 19, 2025 Retrieved Oct 9, 2026 · factual citation
Open source ↗ 37.5
OTIS Mock AIME 2024–2025math 91.4%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] Published Nov 19, 2025 Retrieved Oct 9, 2026 · CC-BY
Open source ↗ 91.4
SWE-bench Pro (public)coding 43.3%Official model cards via models.devLab-reported; metric resolve rate; transcribed by MIT models.dev catalog; not independently evaluated [variant] public Retrieved Oct 9, 2026 · factual citation; MIT transcription
Open source ↗ 43.3 2 settings
  • Setting 1 43.3%Official model cards via models.devLab-reported; metric resolve rate; transcribed by MIT models.dev catalog; not independently evaluated [variant] public Retrieved Oct 9, 2026 · factual citation; MIT transcription
    Open source ↗
  • Setting 2 43.3%SWE-bench Pro (public)Published steward score [variant] Published Nov 26, 2025 Retrieved Oct 9, 2026 · factual citation
    Open source ↗
About SWE-bench Pro (public)
SWE-bench Verifiedcoding 77.4%SWE-bench VerifiedSWE-bench published model plus agent result; harness retained, not a base model evaluation [variant] live-SWE-agentPublished Nov 20, 2025 Retrieved Oct 9, 2026 · factual citation
Open source ↗ 77.4 3 settings
  • Setting 1 72.9%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] Published Feb 13, 2026 Retrieved Oct 9, 2026 · CC-BY
    Open source ↗
  • Setting 2 77.4%SWE-bench VerifiedSWE-bench published model plus agent result; harness retained, not a base model evaluation [variant] live-SWE-agentPublished Nov 20, 2025 Retrieved Oct 9, 2026 · factual citation
    Open source ↗
  • Setting 3 74.2%SWE-bench VerifiedSWE-bench published model plus agent result; harness retained, not a base model evaluation [variant] mini-SWE-agent; 1.15.0Published Nov 18, 2025 Retrieved Oct 9, 2026 · factual citation
    Open source ↗
About SWE-bench Verified

Normalization uses a fixed 0–100 scale for each unit, independent of other models. Compare evaluation conditions before reading a small gap as decisive. “Lab-reported” marks the provider's own published figure.

Details and sources

Open weights
Nomodels.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗
License
not yet reported
Input modalities
text, image, video, audio, pdfmodels.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗
First seen by SuperIndex
Oct 8, 2026
Coverage
54% of expected source weight

Reported (5)

  • Epoch AI BenchmarkingOct 8, 2026
  • Humanity’s Last ExamOct 8, 2026
  • Official model cards via models.devOct 8, 2026
  • SWE-bench VerifiedOct 8, 2026
  • SWE-bench Pro (public)Oct 8, 2026

Awaiting (4)

  • ARC Prize10% of weight
  • LiveBench10% of weight
  • LMArena / Arena20% of weight
  • Terminal-Bench5% of weight

Confidence rises as pending sources publish. Some sources never cover some models, so confidence reaches 100% at 80% of expected weight.

What changed

All releases · RSS feed

Alerts on this device

What to be alerted about
RSS feed