# Google Gemini 3 Flash Preview As of October 22, 2026, Google Gemini 3 Flash Preview ranks #55 of 123 models overall with 55.0 ±10.9 on the LLMBenchmark Index. Gemini 3 Flash Preview ranks #13 of 75 in knowledge, its strongest category. Google Gemini 3 Flash Preview costs $0.50 per million input tokens and $3.00 per million output tokens. Gemini 3 Flash Preview was released on December 17, 2025. Page: https://llmbenchmark.ai/models/gemini-3-flash-preview Data from October 22, 2026. Every figure below names the board it was read from and the day. - Maker: Google - Released: December 17, 2025 - Price per million input tokens: $0.50 - Price per million output tokens: $3.00 - Context window: 1.05M tokens - Overall rank: 55 of 123, by the LLMBenchmark Index - LLMBenchmark Index: 55.0 ±10.9 (90% margin, from 5 of 7 inputs) - Artificial Analysis Intelligence Index: 26.3 - Vibe Coding Index: 43.7 (from 2 of 6 coding benchmarks) Benchmark results (71 of 195): - SWE-bench Verified: 75.8% (source: SWE-bench (swebench.com), read October 22, 2026) - GPQA Diamond: 87.9% (source: GPQA (Vals AI), read October 22, 2026) - AIME: 95.6% (source: AIME (Vals AI), read October 22, 2026) - MMLU Pro: 88.6% (source: MMLU Pro (Vals AI), read October 22, 2026) - AA Intelligence Index: 26.3 (source: Artificial Analysis (via OpenRouter), read October 22, 2026) - Design Arena: websites: 1196 Elo (source: Design Arena, read October 22, 2026) - Design Arena: game dev: 1166 Elo (source: Design Arena, read October 22, 2026) - Design Arena: 3D: 1183 Elo (source: Design Arena, read October 22, 2026) - Design Arena: all code categories: 1189 Elo (source: Design Arena, read October 22, 2026) - Design Arena: agents building web apps: 1111 Elo (source: Design Arena, read October 22, 2026) - Design Arena: agents building full-stack apps: 1063 Elo (source: Design Arena, read October 22, 2026) - Design Arena: agents building mobile apps: 1090 Elo (source: Design Arena, read October 22, 2026) - Design Arena: agents building Android apps: 1051 Elo (source: Design Arena, read October 22, 2026) - Design Arena: agents building Godot games: 1161 Elo (source: Design Arena, read October 22, 2026) - Design Arena: PowerPoint slides: 1036 Elo (source: Design Arena, read October 22, 2026) - LiveCodeBench (Vals AI): 85.6% (source: Vals AI boards, read October 22, 2026) - SWE-bench Verified (Vals AI): 75.0% (source: Vals AI boards, read October 22, 2026) - Vibe Code Bench (Vals AI): 20.2% (source: Vals AI boards, read October 22, 2026) - Code migration (Vals AI): 6.4% (source: Vals AI boards, read October 22, 2026) - ProgramBench (Vals AI): 0.0% (source: Vals AI boards, read October 22, 2026) - Terminal-Bench 2.1 (Vals AI): 53.9% (source: Vals AI boards, read October 22, 2026) - Finance Agent (Vals AI): 42.5% (source: Vals AI boards, read October 22, 2026) - Tax Agent Bench (Vals AI): 11.8% (source: Vals AI boards, read October 22, 2026) - Legal Research Bench (Vals AI): 18.3% (source: Vals AI boards, read October 22, 2026) - Excel Modeling Benchmark (Vals AI): 26.0% (source: Vals AI boards, read October 22, 2026) - MMMU Pro (Vals AI): 87.6% (source: Vals AI boards, read October 22, 2026) - MGSM (Vals AI): 93.3% (source: Vals AI boards, read October 22, 2026) - SAGE (Vals AI): 51.8% (source: Vals AI boards, read October 22, 2026) - LegalBench (Vals AI): 86.9% (source: Vals AI boards, read October 22, 2026) - CaseLaw (Vals AI): 55.8% (source: Vals AI boards, read October 22, 2026) - CorpFin (Vals AI): 66.4% (source: Vals AI boards, read October 22, 2026) - TaxEval (Vals AI): 73.9% (source: Vals AI boards, read October 22, 2026) - MortgageTax (Vals AI): 68.7% (source: Vals AI boards, read October 22, 2026) - MedQA (Vals AI): 95.8% (source: Vals AI boards, read October 22, 2026) - MedCode (Vals AI): 55.9% (source: Vals AI boards, read October 22, 2026) - MedScribe (Vals AI): 69.9% (source: Vals AI boards, read October 22, 2026) - Vals Multimodal Index: 52.2% (source: Vals AI boards, read October 22, 2026) - AA Math Index: 97.0 (source: Artificial Analysis, read October 22, 2026) - GPQA Diamond (Artificial Analysis run): 89.8% (source: Artificial Analysis, read October 22, 2026) - Humanity's Last Exam (Artificial Analysis run): 36.6% (source: Artificial Analysis, read October 22, 2026) - MMLU Pro (Artificial Analysis run): 89.0% (source: Artificial Analysis, read October 22, 2026) - LiveCodeBench (Artificial Analysis run): 90.8% (source: Artificial Analysis, read October 22, 2026) - Terminal-Bench Hard (Artificial Analysis run): 38.6% (source: Artificial Analysis, read October 22, 2026) - IFBench: 78.0% (source: Artificial Analysis, read October 22, 2026) - Long context reasoning (AA-LCR): 78.0% (source: Artificial Analysis, read October 22, 2026) - τ²-Bench Telecom: 80.4% (source: Artificial Analysis, read October 22, 2026) - AIME 2025 (Artificial Analysis run): 97.0% (source: Artificial Analysis, read October 22, 2026) - Terminal-Bench: 64.3% (source: Epoch AI benchmarking hub, read October 22, 2026) - GSO-Bench: 9.8% (source: Epoch AI benchmarking hub, read October 22, 2026) - WeirdML: 61.6% (source: Epoch AI benchmarking hub, read October 22, 2026) - BALROG: 48.1% (source: Epoch AI benchmarking hub, read October 22, 2026) - DeepResearch Bench: 49.8% (source: Epoch AI benchmarking hub, read October 22, 2026) - GPQA Diamond (Epoch AI run): 89.4% (source: Epoch AI benchmarking hub, read October 22, 2026) - ARC-AGI: 84.7% (source: Epoch AI benchmarking hub, read October 22, 2026) - ARC-AGI-2: 33.6% (source: Epoch AI benchmarking hub, read October 22, 2026) - SimpleBench: 61.1% (source: Epoch AI benchmarking hub, read October 22, 2026) - Chess puzzles: 39.0% (source: Epoch AI benchmarking hub, read October 22, 2026) - Mystery game puzzles: 26.0% (source: Epoch AI benchmarking hub, read October 22, 2026) - OTIS Mock AIME: 95.6% (source: Epoch AI benchmarking hub, read October 22, 2026) - FrontierMath, tiers 1 to 3: 51.2% (source: Epoch AI benchmarking hub, read October 22, 2026) - FrontierMath, tier 4: 17.1% (source: Epoch AI benchmarking hub, read October 22, 2026) - ProofBench: 15.0% (source: Epoch AI benchmarking hub, read October 22, 2026) - SimpleQA Verified: 66.8% (source: Epoch AI benchmarking hub, read October 22, 2026) - GeoBench: 88.0% (source: Epoch AI benchmarking hub, read October 22, 2026) - DTBench: 89.1% (source: Epoch AI benchmarking hub, read October 22, 2026) - LMCA: 43.1% (source: Epoch AI benchmarking hub, read October 22, 2026) - VPCT: 72.6% (source: Epoch AI benchmarking hub, read October 22, 2026) - ALE-Bench: 1367.2 (source: ALE-Bench (Sakana AI), read October 22, 2026) - Vending-Bench 2: $3,635 (source: Epoch AI benchmarking hub, read October 22, 2026) - SciCode: 12.7% (source: Kaggle Benchmarks, read October 22, 2026) - FACTS Benchmark Suite: 63.4% (source: Kaggle Benchmarks, read October 22, 2026)