GoogleGemma 3 27B
Best configuration ranks #522 of 610 on the BenchLeader Index at 39.7 ±4.7. Last measured 2 Sept 2026. Released 12 Mar 2025.
- Blended price
- $0.164/M
- $0.119 in · $0.300 out
- Output speed
- 22 tok/s
- First answer
- 1.37 s
- Context
- 131k
- Overall index40
- Reasoning40
- Coding28
- Agents & tools40
- Maths46
- Knowledge32
- Instruction following35
- Human preference54
- Multimodal42
- Long context28
- Composite35
Benchmark results
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 | default | Source |
|---|---|---|
| GPQA Diamond | 47.7%#207 | Epoch AI Benchmarking Hub |
| LMArena Hard Prompts | 1366#178 | LMArena |
| GPQA Diamond (AA)not in index | 42.8%#438 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | 4.4%#426 | Artificial Analysis |
| Kagi LLM Benchmark | 35.1%#117 | Kagi LLM Benchmark |
Coding
| Benchmark | default | Source |
|---|---|---|
| SciCode | 21.2%#153 | SciCode |
| LMArena Coding | 1358#212 | LMArena |
| SciCode (AA)not in index | 23.3%#156 | Artificial Analysis |
| SWE-Bench Pro | 11.4%#20 | Scale AI SEAL |
| Aider Polyglot | 4.9%#44 | Aider polyglot leaderboard |
Agents & tools
| Benchmark | default | Source |
|---|---|---|
| Terminal-Bench Hard | 3.8%#283 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | 10.5%#366 | Artificial Analysis |
| BFCL Overall | 29.5%#51 | Berkeley Function Calling Leaderboard |
Maths
| Benchmark | default | Source |
|---|---|---|
| OTIS Mock AIME | 22.5%#200 | Epoch AI Benchmarking Hub |
| MATH Level 5 | 74.0%#35 | Epoch AI Benchmarking Hub |
Knowledge
| Benchmark | default | Source |
|---|---|---|
| AA-Omniscience | -67.2#428 | Artificial Analysis |
Instruction following
| Benchmark | default | Source |
|---|---|---|
| IFBench | 31.8%#331 | Artificial Analysis |
Human preference
| Benchmark | default | Source |
|---|---|---|
| LMArena Text | 1365#164 | LMArena |
Multimodal
| Benchmark | default | Source |
|---|---|---|
| LMArena Vision | 1164#91 | LMArena |
| MMMU-Pro | 48.0%#213 | Artificial Analysis |
| MMMU (validation) | 64.9%#27 | MMMU |
Long context
| Benchmark | default | Source |
|---|---|---|
| AA-LCR | 7.3%#403 | Artificial Analysis |
Composite
| Benchmark | default | Source |
|---|---|---|
| Epoch Capabilities Indexnot in index | 130.0#132 | Epoch AI Benchmarking Hub |
| AA Intelligence Index | 4.8#525 | Artificial Analysis |
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 |
|---|---|---|---|---|---|---|
| Nebius Token Factory | 29 tok/s | 1.21 s | $0.100 | $0.300 | 110k | fp8 |
| Parasail | 26 tok/s | 1.30 s | $0.080 | $0.450 | 131k | fp8 |
| NovitaAI | 17 tok/s | 1.44 s | $0.119 | $0.200 | 98k | bf16 |
| DeepInfra | 13 tok/s | 2.63 s | $0.080 | $0.160 | 131k | fp8 |
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 $0.010 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0001 | $0.0001 | 15.3 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0016 | $0.0006 | 29.3 s |
| Code edit | 6,000 / 1,500 | $0.0012 | $0.0007 | 1.2 min |
| Agentic coding session | 60,000 / 4,000 | $0.0083 | $0.0034 | 3.1 min |
| Structured extraction | 2,000 / 200 | $0.0003 | $0.0001 | 10.7 s |
See also
Data as of 9 Sept 2026. Compare with another model.