GoogleGemini 3 Flash
Gemini 3 Flash is a Google proprietary reasoning model, released 17 Dec 2025. Its best configuration (thinking reasoning effort) ranks #138 of 785 on the BenchLeader Index at 59.1 ±5.3, in the upper half. The ± is the point: 248 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in instruction following (76) and lowest in reasoning (55). At $1.13 per million tokens blended it is pricier than most ranked models. Output speed of 222 tokens per second puts it in the fastest quarter, with a first answer in 6.0 s. It has been measured at 7 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 17 Dec 2025.stale: no new result in six months
- Blended price
- $1.13/M
- $0.500 in · $3.00 out
- Output speed
- 222 tok/s
- measured by Artificial Analysis
- First answer
- 5.97 s
- first token 1.25 s
- Context
- 1.0M
- Full answer
- 8.22 s
- median, reasoning included
- Cost per run
- –
- Released
- 17 Dec 2025
- Overall index59
- Reasoning55
- Agents & tools55
- Knowledge66
- Instruction following76
- Multimodal63
- Long context64
- Composite59
Versions
Google has shipped 12 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.
| Model | Released | Index | Rank |
|---|---|---|---|
| Gemini 3.8 Flashhigh | 2 Sept 2026 | 64.4 | #50 |
| Gemini 3.7 Flashhigh | 13 Aug 2026 | 63.5 | #64 |
| Gemini 3.6 Flashhigh | 21 Jul 2026 | 62.0 | #89 |
| Gemini 3.5 Flashhigh | 19 May 2026 | 63.1 | #70 |
| Gemini 3 Flashthinkingthis page | 17 Dec 2025 | 59.1 | #138 |
| Gemini 2.5 Flash 09 2025thinking | 25 Sept 2025 | 51.1 | #351 |
| Gemini 2.5 Flash | 17 Apr 2025 | 52.7 | #306 |
| Gemini 2.0 Flash | 5 Feb 2025 | 43.2 | #574 |
| Gemini 1.5 Flash 002 | 24 Sept 2024 | 38.3 | #716 |
| Gemini 1.5 Flash 001 | 23 May 2024 | – | – |
| Gemini 1.5 Flash | 14 May 2024 | 44.3 | #541 |
| Gemini Flash 2.0 | – | – | – |
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| no reasoning | 53.3 | #292 | 211 tok/s | 0.92 s | $0.0011 | Agents & tools 62 · Composite 50 · Instruction following 56 · Knowledge 60 · Long context 52 · Multimodal 62 · Reasoning 43 |
| minimal | 54.0 | #262 | 109 tok/s | 1.25 s | $0.0011 | Coding 49 · Human preference 65 · Maths 63 · Multimodal 60 · Reasoning 41 |
| low | – | – | 109 tok/s | 1.25 s | $0.0011 | Reasoning 33 |
| medium | – | – | 109 tok/s | 1.25 s | $0.0011 | Reasoning 41 |
| high | 58.3 | #154 | 109 tok/s | 1.25 s | $0.0011 | Agents & tools 52 · Coding 64 · Knowledge 64 · Maths 63 · Reasoning 57 |
| thinkingbest | 59.1 | #138 | 222 tok/s | 5.97 s | $0.0011 | Agents & tools 55 · Composite 59 · Instruction following 76 · Knowledge 66 · Long context 64 · Multimodal 63 · Reasoning 55 |
| not stated | 57.0 | #187 | 109 tok/s | 1.25 s | $0.0011 | Agents & tools 52 · Coding 52 · Human preference 66 · Maths 54 · Multimodal 63 · Reasoning 61 |
How its index has moved
8 Sept 2026 to 23 Oct 2026- thinking
- high
- not stated
- minimal
The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.
Benchmark results
See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | no reasoning | minimal | low | medium | high | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|---|
| GPQA Diamond | – | – | – | – | 89.4%#53 | – | 83.2%#109 | Epoch AI Benchmarking Hub | |
| SimpleBencheffort not stated | – | – | – | – | – | – | 61.1%#26 | SimpleBench | |
| Arena Hard Prompts | – | 1474#77 | – | – | – | – | 1493#45 | Arena (LMArena) | |
| GPQA Diamond (AA)not in index | 81.2%#182 | – | – | – | – | 89.8%#68 | – | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 15.0%#253 | – | – | – | – | 36.6%#111 | – | Artificial Analysis | |
| GPQA Diamond (Vals)not in index | – | – | – | – | 87.9%#36 | – | – | Vals AI | |
| ARC-AGI-1 | – | 21.5%#218 | 29.0%#209 | 57.7%#166 | 84.7%#113 | – | – | ARC Prize | |
| ARC-AGI-2 | – | 3.3%#192 | 1.3%#218 | 12.8%#155 | 33.6%#123 | – | – | ARC Prize | |
| CritPt | 1.4%#211 | – | – | – | – | 8.6%#134 | – | Artificial Analysis | |
| SciPredictnot in indexeffort not stated | – | – | – | – | – | – | 22.2%#4 | Scale AI SEAL | |
| Chess Puzzlesnot in index | – | – | – | – | 39.0%#43 | – | – | Epoch AI Benchmarking Hub | |
| Mystery Game Puzzlesnot in index | – | 25.0%#47 | 26.0%#43 | – | 20.0%#60 | – | – | Epoch AI Benchmarking Hub | |
| LMCAnot in index | – | – | – | – | 43.1%#100 | – | – | Epoch AI Benchmarking Hub | |
| DTBenchnot in index | – | – | – | – | 89.1%#92 | – | – | Epoch AI Benchmarking Hub | |
| ForecastBenchnot in indexeffort not stated | – | – | – | – | – | – | 64.7%#21 | ForecastBench |
Coding
Agents & tools
| Benchmark | no reasoning | minimal | low | high | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| Terminal-Bencheffort not stated | – | – | – | – | – | 64.3%#12 | Terminal-Bench | |
| Terminal-Bench Hard | 31.8%#101 | – | – | – | 38.6%#60 | – | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 43.3%#216 | – | – | – | 80.4%#123 | – | Artificial Analysis | |
| Terminal-Bench 2.1 (Vals) | – | – | – | 53.9%#51 | – | – | Vals AI | |
| MCP Atlaseffort not stated | – | – | – | – | – | 62.0%#26 | Scale AI SEAL | |
| τ²-bench | – | – | – | 83.5%#4 | – | – | τ²-bench | |
| τ³-bench Bankingnot in index | – | – | – | 27.3%#14 | – | – | τ²-bench | |
| τ³-Banking (AA) | – | – | – | – | 20.8%#94 | – | Artificial Analysis | |
| APEX-Agents (AA)not in index | – | – | – | – | 27.7%#16 | – | Artificial Analysis | |
| Finance Agent v2not in index | – | – | – | 42.5%#58 | – | – | Vals AI | |
| Harvey's Legal Agent Benchmark v1not in index | – | – | – | 0.0%#61 | – | – | Vals AI | |
| Legal Research Benchnot in index | – | – | – | 18.3%#59 | – | – | Vals AI | |
| Poker Agentnot in indexeffort not stated | – | – | – | – | – | 1100.2%#3 | Vals AI | |
| Tax Agent Benchnot in index | – | – | – | 11.8%#55 | – | – | Vals AI | |
| Terminal-Bench 2.0 (Vals)not in index | – | – | – | 51.7%#18 | – | – | Vals AI | |
| BALROGnot in indexeffort not stated | – | – | – | – | – | 48.1%#7 | BALROG | |
| DeepResearch Benchnot in index | – | 49.0%#15 | 49.8%#11 | 47.9%#21 | – | – | DeepResearch Bench | |
| Vending-Bench 2not in indexeffort not stated | – | – | – | – | – | 3634.7#45 | Andon Labs | |
| BrowseCompnot in indexeffort not stated | – | – | – | – | – | 16.9%#3 | OpenAI | |
| FACTS Searchnot in indexeffort not stated | – | – | – | – | – | 80.9%#11 |
Maths
| Benchmark | minimal | high | not stated | Source | Trend |
|---|---|---|---|---|---|
| FrontierMath Tiers 1–3effort not stated | – | – | 51.2%#53 | Epoch AI Benchmarking Hub | |
| FrontierMath Tier 4effort not stated | – | – | 17.1%#56 | Epoch AI Benchmarking Hub | |
| OTIS Mock AIME | – | 95.6%#44 | 92.8%#65 | Epoch AI Benchmarking Hub | |
| ProofBencheffort not stated | – | – | 15.0%#59 | Vals AI | |
| AIME (Vals) | – | 95.6%#8 | – | Vals AI | |
| MGSM | – | 93.3%#10 | – | Vals AI | |
| AIME 2026effort not stated | – | – | 96.7%#6 | MathArena | |
| HMMT February 2026effort not stated | – | – | 89.4%#14 | MathArena | |
| MathArena Apexeffort not statedpossible contamination | – | – | 15.6%#16 | MathArena | |
| Arena Maths | 1454#70 | – | 1475#46 | Arena (LMArena) |
Knowledge
| Benchmark | no reasoning | high | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| SimpleQA Verified | – | 66.8%#11 | – | – | Epoch AI Benchmarking Hub | |
| AA-Omniscience | -4.3#158 | – | 10.1#96 | – | Artificial Analysis | |
| MMLU-Pro | – | 88.6%#19 | – | – | Vals AI | |
| LegalBench | – | 86.9%#11 | – | – | Vals AI | |
| CorpFin | – | 66.4%#19 | – | – | Vals AI | |
| TaxEval | – | 73.9%#44 | – | – | Vals AI | |
| MedQA | – | 95.8%#11 | – | – | Vals AI | |
| CaseLaw v2not in index | – | 55.8%#32 | – | – | Vals AI | |
| Excel Modeling Benchmarknot in index | – | 26.0%#65 | – | – | Vals AI | |
| MedCodenot in index | – | 55.9%#5 | – | – | Vals AI | |
| MedScribenot in index | – | 69.9%#91 | – | – | Vals AI | |
| GDP.pdfnot in index | – | 10.0%#50 | – | 10.0%#50 | Epoch AI Benchmarking Hub | |
| AA-Omniscience: accuracynot in index | 45.8%#84 | – | 53.4%#48 | – | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | 7.6%#480 | – | 7.0%#492 | – | Artificial Analysis | |
| Global-MMLU-Lite (AA)not in index | 92.8%#2 | – | 91.4%#5 | – | Artificial Analysis |
Instruction following
| Benchmark | no reasoning | minimal | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| IFBench | 55.1%#148 | – | 78.0%#14 | – | Artificial Analysis | |
| Arena Instruction Followingnot in index | – | 1443#76 | – | 1458#53 | Arena (LMArena) |
Human preference
| Benchmark | minimal | not stated | Source | Trend |
|---|---|---|---|---|
| Arena Text | 1458#63 | 1473#44 | Arena (LMArena) | |
| Arena Creative Writingnot in index | 1447#44 | 1457#30 | Arena (LMArena) | |
| Arena Multi-turnnot in index | 1474#51 | 1482#37 | Arena (LMArena) | |
| Arena Longer Queriesnot in index | 1462#72 | 1476#53 | Arena (LMArena) |
Multimodal
| Benchmark | no reasoning | minimal | high | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|
| Arena Vision | – | 1266#48 | – | – | 1285#32 | Arena (LMArena) | |
| MMMU-Pro | 78.5%#69 | – | – | 79.9%#54 | – | Artificial Analysis | |
| MMMU-Pro (Vals)not in index | – | – | 87.6%#13 | – | – | Vals AI | |
| MortgageTaxnot in index | – | – | 68.7%#11 | – | – | Vals AI | |
| SAGEnot in index | – | – | 51.9%#10 | – | – | Vals AI | |
| GeoBenchnot in indexeffort not stated | – | – | – | – | 88.0%#1 | GeoBench | |
| VPCTnot in indexeffort not stated | – | – | – | – | 72.6%#3 | Epoch AI Benchmarking Hub | |
| Blueprint-Bench 2not in indexeffort not stated | – | – | – | – | 0.0%#29 | Epoch AI Benchmarking Hub |
Long context
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| AA-LCR | 55.3%#298 | 78.0%#121 | – | Artificial Analysis | |
| Arena Documentnot in indexeffort not stated | – | – | 1439#36 | Arena (LMArena) |
Composite
| Benchmark | no reasoning | high | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| Epoch Capabilities Indexnot in indexeffort not stated | – | – | – | 151.9#52 | Epoch AI Benchmarking Hub | |
| AA Intelligence Index v4.3.2 | 17.9#259 | – | 26.3#155 | – | Artificial Analysis | |
| Vals Multimodal Indexnot in index | – | 52.2%#23 | – | – | Vals AI |
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| Google Vertex | 124 tok/s | 1.13 s | $0.900 | $5.40 | 1.0M | – |
| Google Vertex | 111 tok/s | 1.20 s | $0.500 | $3.00 | 1.0M | – |
| Google AI Studio | 108 tok/s | 1.50 s | $0.250 | $1.50 | 1.0M | – |
| Google AI Studio | 85 tok/s | 1.55 s | $0.500 | $3.00 | 1.0M | – |
| Google AI Studio | 77 tok/s | 1.60 s | $0.900 | $5.40 | 1.0M | – |
| Google Vertex | 47 tok/s | 14 s | $0.250 | $1.50 | 1.0M | – |
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache at $0.050 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0011 | $0.0010 | 7.3 s |
| Summarize a 30-page report | 12,000 / 600 | $0.0078 | $0.0038 | 8.7 s |
| Code edit | 6,000 / 1,500 | $0.0075 | $0.0055 | 12.7 s |
| Agentic coding session | 60,000 / 4,000 | $0.042 | $0.022 | 24.0 s |
| Structured extraction | 2,000 / 200 | $0.0016 | $0.0009 | 6.9 s |
More from Google
See also
Questions people ask
- What are Gemini 3 Flash's benchmark scores?
- As of 24 Oct 2026, Gemini 3 Flash (thinking effort) scores 59.1 ±5.3 on the BenchLeader Index, strongest in instruction following (76) and weakest in reasoning (55). That rests on 8 independent benchmark results, and every score on this page links to its publisher.
- How does Gemini 3 Flash rank?
- Gemini 3 Flash (thinking effort) is #138 of 785 ranked configurations on 24 Oct 2026. 248 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Opus 5.5 (max effort) at 72.6.
- How much does Gemini 3 Flash cost per 1M tokens?
- As of 24 Oct 2026, Gemini 3 Flash costs $0.500 per 1M input tokens and $3.00 per 1M output tokens, $1.13 blended (three input tokens per output token).
- When was Gemini 3 Flash released?
- Google released Gemini 3 Flash on 17 Dec 2025. BenchLeader has results for it at 7 reasoning-effort settings.
Data as of 24 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “Gemini 3 Flash: benchmarks, pricing, speed and rank”, data as of 24 Oct 2026, https://www.benchleader.com/models/gemini-3-flash.