Compare / head-to-head
Gemini 3 Pro PreviewvsGemini 4 Argon
Gemini 4 Argon leads 2 of 2 shared benchmarks. Gemini 4 Argon is 1.2x cheaper per token.
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
0 – 2
Gemini 4 Argon leads
Cheaper per token
Gemini 4 Argon
1.2x cheaper, input + output
Larger context
—
both 1.0M tokens
Providers
Google
same provider
Gemini 3 Pro Preview
Google · released 2025-11-18
textvisionaudiovideo
Gemini 4 Argon
Google · released 2026-09-30
textvisionvideo
Quality
Benchmark matrix
2 shared · 22 only Gemini 3 Pro Preview · 10 only Gemini 4 Argon| Benchmark | Gemini 3 Pro Preview | Gemini 4 Argon | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 2 | ||||
| SAGE (Vals AI) | 47.6% ↗ | 53.6% ↗ | -6 pt | Gemini 4 Argon |
| Vending-Bench 2 (Andon Labs) | 5478.16 ↗ | 13718.16 ↗ | -8240 | Gemini 4 Argon |
| Only Gemini 3 Pro Preview reports · 22 | ||||
| SWE-bench Verified | 76.2% ↗ | not reported | — | — |
| GPQA Diamond | 91.9% ↗ | not reported | — | — |
| AIME 2025 | 95% ↗ | not reported | — | — |
| Humanity's Last Exam (no tools) | 37.5% ↗ | not reported | — | — |
| AIME 2025 (with code execution) | 100% ↗ | not reported | — | — |
| ARC-AGI-2 (ARC Prize Verified) | 31.1% ↗ | not reported | — | — |
| BALROG (BALROG) | 58.1% ↗ | not reported | — | — |
| CharXiv Reasoning | 81.4% ↗ | not reported | — | — |
| Chess Puzzles (Epoch AI run) | 31% ↗epoch run | not reported | — | — |
| GSO Opt@1 (GSO) | 17.7% ↗ | not reported | — | — |
| Humanity's Last Exam (with search and code execution) | 45.8% ↗ | not reported | — | — |
| MathArena Apex | 23.4% ↗ | not reported | — | — |
| MMMLU | 91.8% ↗ | not reported | — | — |
| MMMU-Pro | 81% ↗ | not reported | — | — |
| OTIS Mock AIME 2024-2025 (Epoch AI run) | 91.4% ↗epoch run | not reported | — | — |
| SimpleBench (SimpleBench) | 76.4% ↗ | not reported | — | — |
| SimpleQA Verified | 72.1% ↗ | not reported | — | — |
| SWE-bench Verified (Epoch AI run) | 72.9% ↗epoch run | not reported | — | — |
| Terminal-Bench 2.0 (Terminus-2 agent) | 54.2% ↗ | not reported | — | — |
| Video-MMMU | 87.6% ↗ | not reported | — | — |
| WeirdML (Håvard Tveit Ihle) | 69.9% ↗ | not reported | — | — |
| τ2-bench | 85.4% ↗ | not reported | — | — |
| Only Gemini 4 Argon reports · 10 | ||||
| LMArena Elo | not reported | 1525.2 ↗ | — | — |
| APEX-Agents (Mercor) | not reported | 82.2% ↗ | — | — |
| AutomationBench | not reported | 51.3% ↗ | — | — |
| CWE-bench v1 | not reported | 68% ↗ | — | — |
| DeepSWE v1.1 | not reported | 77.9% ↗ | — | — |
| FrontierSWE V2 (Proximal Labs) | not reported | 55% ↗ | — | — |
| LMArena Agent (LMArena) | not reported | 0.0757 ↗ | — | — |
| LMArena WebDev (LMArena) | not reported | 1679.5 ↗ | — | — |
| LVBench | not reported | 91.7% ↗ | — | — |
| Terminal-Bench 4.0 (Vals AI) | not reported | 57.6% ↗ | — | — |
Scores tagged "epoch" or "matharena" are independent runs, used only where the lab has not published its own; ⚠ marks rows MathArena flags as released after the competition. Higher is better on every row. Δ is Gemini 3 Pro Preview minus Gemini 4 Argon in the benchmark's own unit. "Not reported" means the lab has not published that figure; it is not a zero. ↗ opens the source.
Specs & pricing
Side by side
Official pricing pages and model cards| Spec | Gemini 3 Pro Preview | Gemini 4 Argon | Edge |
|---|---|---|---|
| Context window | 1.0M tokens | — | — |
| Max output | 66K tokens | 1M tokens | Gemini 4 Argon |
| Input price / 1M | $2 ↗ | $2 ↗ | Tie |
| Output price / 1M | $12 ↗ | $10 ↗ | Gemini 4 Argon |
| Cached input / 1M | — | $0.1 ↗ | — |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | $14 | $12 | Gemini 4 Argon |
| Modalities | text · vision · audio · video | text · vision · video | Gemini 3 Pro Preview |
| Released | 2025-11-18 | 2026-09-30 | — |
| Cited benchmark scores | 25 | 12 | — |
Reliability
Google status
Live from /statusMore matchups
Gemini 3 Pro Preview vs …
Models sharing the most benchmarksMore matchups
Gemini 4 Argon vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run Gemini 3 Pro Preview and Gemini 4 Argon on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.