Side-by-side comparison · models
Gemini 2.5 Pro vs Gemini 3 Flash Preview
What we have checked about each, line by line. Every value shows where it came from and when we last confirmed it.
Models
Gemini 2.5 Pro
Gemini 2.5 Pro is Google’s state-of-the-art AI model designed for advanced reasoning, coding, mathematics, and scientific tasks.
Models
Gemini 3 Flash Preview
Gemini 3 Flash Preview is a high speed, high value thinking model designed for agentic workflows, multi turn chat, and coding assistance.
Checked facts
Line by line
| Measure | Gemini 2.5 Pro | Gemini 3 Flash Preview |
|---|---|---|
| Input price | $1.25 per million tokens | $0.500 per million tokens |
| Output price | $10.00 per million tokens | $3.00 per million tokens |
| Price on google-ai-studio | $0.063 per million cached input tokens | $0.025 per million cached input tokens |
| Price on google-ai-studio | $0.625 per million input tokens | $0.250 per million input tokens |
| Price on google-ai-studio | $5.00 per million output tokens | $1.50 per million output tokens |
| Price on google-vertex | Not checked yet | $0.025 per million cached input tokens |
| Price on google-vertex | Not checked yet | $0.250 per million input tokens |
| Price on google-vertex | Not checked yet | $1.50 per million output tokens |
| Latest version | google/gemini-2.5-pro | google/gemini-3-flash-preview-20251217 |
| Time to first token (ms) | 22,230 | 810 |
| Time to first token (ms) on google-ai-studio | 3,475 | 1,996 |
| Output speed (tokens a second) | 114 | 208.8 |
| Output speed (tokens a second) on google-ai-studio | 70 | 93 |
| Time to first token (ms) on google-vertex | Not checked yet | 13,935 |
| Output speed (tokens a second) on google-vertex | Not checked yet | 14 |
| Tokens served on OpenRouter in a day | 4.3B (for 10 Nov 2026) | 94.4B (for 10 Nov 2026) |
| Tool calls on OpenRouter in a day | 10.2K (for 10 Nov 2026) | 1.1M (for 10 Nov 2026) |
| Artificial Analysis Intelligence Index | 16.1 | 26.3 |
| Artificial Analysis Coding Index | 33.3 | Not checked yet |
| Artificial Analysis Agentic Index | 1.6 | Not checked yet |
| LMArena text rating (Elo) | 1,458 | Not checked yet |
| LMArena web development rating (Elo) | 1,228 | Not checked yet |
| LMArena vision rating (Elo) | 1,263 | Not checked yet |
| Design Arena rating (Elo) | 1,220 | 1,192 |
| Gpqa diamond | 85.3% (for 17 June 2025) | 89.4% (for 17 Dec 2025) |
| Swe bench verified | 57.6% (for 17 June 2025) | 75.4% (for 17 Dec 2025) |
| Frontiermath tiers 1 3 v2 private | 24.6% (for 17 June 2025) | 51.2% (for 17 Dec 2025) |
| Terminal bench | 32.6% (for 17 June 2025) | 64.3% (for 17 Dec 2025) |
| Arc agi 2 | 4.9% (for 17 June 2025) | 33.6% (for 17 Dec 2025) |
| Arc agi | 41% (for 17 June 2025) | 84.7% (for 17 Dec 2025) |
| Deepresearch bench | 41.5% (for 17 June 2025) | 49.8% (for 17 Dec 2025) |
| Dtbench | 82.4% (for 17 June 2025) | 89.1% (for 17 Dec 2025) |
| Frontiermath 2025 02 28 private | 14.1% (for 17 June 2025) | 35.6% (for 17 Dec 2025) |
| Frontiermath tier 4 2025 07 01 private | 4.2% (for 17 June 2025) | 4.2% (for 17 Dec 2025) |
| Frontiermath tier 4 v2 private | 0% (for 17 June 2025) | 17.1% (for 17 Dec 2025) |
| Gdpval | 23.3% (for 17 June 2025) | Not checked yet |
| Gso bench | 3.9% (for 17 June 2025) | 9.8% (for 17 Dec 2025) |
| Lech mazur writing | 83.8% (for 17 June 2025) | Not checked yet |
| Lmca | 34.8% (for 17 June 2025) | 43.1% (for 17 Dec 2025) |
| Weirdml | 54% (for 17 June 2025) | 61.6% (for 17 Dec 2025) |
| Otis mock aime 2024 2025 | 84.7% (saturated benchmark: the top models all score near the ceiling) (for 17 June 2025) | 95.6% (saturated benchmark: the top models all score near the ceiling) (for 17 Dec 2025) |
| Simpleqa verified | Not checked yet | 66.8% (for 17 Dec 2025) |
| Balrog | Not checked yet | 48.1% (for 17 Dec 2025) |
| Chess puzzles | Not checked yet | 39% (for 17 Dec 2025) |
| Geobench | Not checked yet | 88% (for 17 Dec 2025) |
| Mystery game puzzles | Not checked yet | 26% (for 17 Dec 2025) |
| Proofbench | Not checked yet | 15% (for 17 Dec 2025) |
| Simplebench | Not checked yet | 61.1% (for 17 Dec 2025) |
| Vpct | Not checked yet | 72.6% (for 17 Dec 2025) |
A line appears only when at least one record has a value we confirmed from a source. Values we have not confirmed are left out, never guessed. Records of different types share fewer lines.
Add a third
Compare with one more record
How this works
Share it, or build your own
This page works for any two or three records in the directory, of any type. The link stays the same whichever order you name them in, so you can share it as it is. Change the records in the address, for example /directories/vs/claude-code-vs-cursor, or start from our editorial comparisons.