Reasoning-first Gemini preview for agentic coding and complex problem solving
Gemini 3.1 Pro Preview by google is offered by 35 providers on this page. Public prices are shown for 33 of them. Its official list price is $2.00 per 1M input tokens and $12.00 per 1M output tokens. The lowest paid channel is Kilo Gateway at $1.00 / $6.00 per 1M, about 2.5× below the list price. That channel is a third-party gateway, so confirm its availability and rate limits before depending on it. It also has 1 free ($0) channel; free tiers usually carry rate limits, and subscription-covered access bills $0 per token only after the subscription fee.
Artificial Analysis rates it 29.7 on the Intelligence Index, with 68.8 for coding and 8.2 for agentic tasks. Against its lowest blended price of $2.25 per 1M, that is roughly 13 index points per dollar, which is the value ratio the leaderboards rank on. Median output speed is 121 tokens per second, with 24.15s to the first token. Running one task of the Artificial Analysis suite costs about $0.6747, which reflects how many tokens its reasoning consumes rather than the unit price alone.
The context window is 1,048,576 tokens at the reference host, but hosts report different limits, from 1,000,000 to 1,048,576, so the usable window depends on the provider you pick. It supports reasoning, tool use, and structured output. Accepted input modalities are Text, Image, Video, Audio, and PDF. Its training knowledge cuts off at 2025-01.
Quality is independently evaluated by Artificial Analysis. Speed/latency are model-level medians.
| Provider | Tier | Input | Output | Cache read | Cache write | Context | Output limit | Status |
|---|---|---|---|---|---|---|---|---|
| Kenari gemini-3-1-pro | Gateway | Free | — | — | 1,048,576 | 65,536 | ||
| Kilo Gateway | Gateway | $1.00 | $6.00 | $0.10 | $0.19 | 1,048,576 | 65,536 | |
| Perplexity Agent | Gateway | $2.00 tiered >200K: $4.00 | $12.00 | $0.20 | — | 1,048,576 | 65,536 | |
| NanoGPT | Gateway | $2.00 | $12.00 | $0.20 | $0.38 | 1,048,576 | 65,536 | |
| Abacus | Gateway | $2.00 | $12.00 | $0.20 | — | 1,048,576 | 65,536 | |
| 302.AI | Gateway | $2.00 | $12.00 | — | — | 1,048,576 | 65,536 | |
| OpenRouter | Gateway | $2.00 tiered >200K: $4.00 | $12.00 | $0.20 | $0.38 | 1,048,576 | 65,536 | |
| Auriko | Gateway | $2.00 tiered >200K: $4.00 | $12.00 | $0.20 | — | 1,048,576 | 65,536 | |
| ZenMux | Gateway | $2.00 | $12.00 | $0.20 | $4.50 | 1,048,000 ⚠ | 64,000 | |
| Impossibl | Gateway | $2.00 tiered >200K: $4.00 | $12.00 | $0.20 | — | 1,048,576 | 65,536 | |
| GoogleOfficial | First-party | $2.00 tiered >200K: $4.00 | $12.00 | $0.20 | — | 1,048,576 | 65,536 | |
Sorted by blended price (input×0.75 + output×0.25) asc. The official channel always shows regardless of rank. Whether a gateway's low price is actually usable can't be verified.
Interleaved thinking (reasoning between tool calls) is declared by 3 of 35 providers.
3 / 35 providers expose no reasoning control (reasoning_options: []).
| Provider | Audio in | Audio out |
|---|---|---|
| OrcaRouter | $2.001.0× | — |
| Eden AI | $2.001.0× | — |
| Eden AI | $2.001.0× | — |
Audio tokens are priced on a separate scale from text tokens, so these are not comparable with the text prices above. The multiplier compares each price with the same provider's text price for the same direction.
1 more channels offer $0 (Kenari); free tiers usually have rate limits and no SLA, excluded from ranking.
| Name | Conditions | Score | Metric | Source |
|---|---|---|---|---|
| SWE-Bench Pro | — | 54.2 | resolve rate | Source ↗ |
| Terminal-Bench | harness: Terminus-2 · v2.1 | 70.3 | success rate | Source ↗ |
| SWE-Bench Pro | dataset: public | 46.1 | resolve rate | Source ↗ |
| SWE-Atlas Codebase QnA | harness: Mini-SWE-Agent | 13.5 | score | Source ↗ |
| SWE-Atlas Refactoring | harness: Gemini CLI | 33.81 | score | Source ↗ |
| SWE-Atlas Test Writing | harness: Mini-SWE-Agent | 29.84 | score | Source ↗ |
| Artificial Analysis Coding Agent Index | harness: Gemini CLI · variant: high | 43 | average pass@1 | Source ↗ |
| SWE-Atlas Codebase QnA | harness: Gemini CLI · variant: high | 45.6 | pass@1 | Source ↗ |
| SWE-Bench Pro | harness: Gemini CLI · variant: high · dataset: hard-aa | 15.1 | pass@1 | Source ↗ |
| Terminal-Bench | harness: Gemini CLI · variant: high · v2.1 | 68.3 | pass@1 | Source ↗ |
| GPQA Diamond | — | 94.3 | accuracy | Source ↗ |
| Humanity's Last Exam | dataset: full set, text + MM | 44.4 | accuracy | Source ↗ |
| ARC-AGI-2 | — | 77.1 | accuracy | Source ↗ |
| MMMU Pro | variant: no tools | 80.5 | accuracy | Source ↗ |
| MCP Atlas | — | 78.2 | success rate | Source ↗ |
| OSWorld-Verified | — | 76.2 | success rate | Source ↗ |
| CharXiv Reasoning | variant: no tools | 83.3 | accuracy | Source ↗ |
| GDPval-AA | — | 1314 | Elo | Source ↗ |
The same benchmark scores very differently across harness / dataset, so the qualifying conditions must be shown together.
Individual evaluations run by Artificial Analysis, on the same reasoning tier as the intelligence score above. Each benchmark has its own task set and harness, so rows are not comparable with one another. The Intelligence Index above draws on a different, newer set of evaluations.