Gemini 3.1 Pro · Media / benchmark · Independent measurement
Artificial Analysis compares 182 similarly priced models on first-party APIs and reports Gemini 3.1 Pro Preview at Intelligence Index 48, 121.4 t/s, and 32.45 seconds TTFT, combining high throughput with high startup latency.
Independent testing by Artificial Analysis shows that Gemini 3.1 Pro Preview achieves an Intelligence Index score of 48 (vs. a price-tier median of 35) with an output generation speed of 121.4 t/s (vs. a price-tier median of 76.2 t/s); however, driven by its extended reasoning chain, its time to first token (TTFT) reaches 32.45s, making it a classic reasoning model characterized by high intelligence, high throughput, and high upfront reasoning latency.
Evaluation organization: Artificial Analysis independent benchmark platform.
Evaluation target: gemini-3.1-pro-preview (Google official first-party API endpoint).
Comparison baseline: 182 mainstream large language models within the same price tier (>$1.00/1M tokens).
Evaluation framework: Artificial Analysis Intelligence Index v4.1.1 (aggregating 9 benchmarks: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, and AA-LCR).
Test configuration: Standardized on first-party API default parameters and reasoning settings, measuring end-to-end token consumption, output generation throughput, time to first token (TTFT), and API invocation costs.
Input modalities: Text, image, audio, and video.
Output modality: Text.
Context window: 1,000,000 tokens (1M).
API pricing: $2.00 / 1M tokens for input, $12.00 / 1M tokens for output, with a 90% discount on context cache hits (Cache Hit: $0.20 / 1M tokens).
| Evaluation Metric | Gemini 3.1 Pro Measured | Price-Tier Median / Rank | Key Observations & Characteristics |
|---|---|---|---|
| Intelligence Index (Overall Intelligence) | 48 | 35 (Rank #40 / 182) | Significantly above the peer median, ranking in the top tier |
| Output Speed (Generation Throughput) | 121.4 t/s | 76.2 t/s (Rank #35 / 182) | Extremely fast decoding generation, well-suited for batch throughput |
| Time to First Token (TTFT) | 32.45s | 2.79s | Includes deep "thinking" phase, resulting in longer upfront waiting time |
| Total Generated Tokens (Verbosity) | 56M | 72M (Rank #30 / 182) | More concise with lower verbosity compared to peer reasoning models |
| Weighted Cost per Task | $0.33 | Mid-to-high tier | Driven by the $12/1M output pricing combined with multi-step reasoning |
| Total Evaluation Cost (Full Benchmark Suite) | $811.90 | - | Standard full evaluation cost covering all 9 sub-benchmarks |
In independent controlled benchmarks, Gemini 3.1 Pro Preview demonstrates a standout profile of "high-speed decoding + concise output" (121.4 t/s, with significantly fewer generated tokens than the peer median), making it well-suited for long-document reasoning and high-throughput analytical workloads. However, its 32.45s TTFT makes it unsuitable for low-latency real-time conversational applications or interactive UIs that require sub-second initial responses.
Testing is based on preview snapshots from the official Google API; performance in production environments may experience fluctuations due to server load variance.
The Intelligence Index does not isolate or break down the marginal performance-latency tradeoffs of different thinking_level configurations (low/medium/high) across individual academic sub-benchmarks.
Evaluations were conducted exclusively via the first-party API, without accounting for routing overhead introduced by third-party distribution platforms (e.g., OpenRouter, Vertex AI proxies).
Access the Google GenAI API and specify gemini-3.1-pro-preview as the model.
Execute the evaluation scripts for the 9 open-source or controlled sub-datasets corresponding to the Artificial Analysis Intelligence Index v4.1.1.
Enable streaming to record the duration from request dispatch to receipt of the first answer token (TTFT) and the subsequent token generation throughput (tokens per second).
Aggregate input and output token counts, and calculate actual per-task costs based on official API pricing.
The figures, task set, reasoning tier, and client conditions apply only to the listed source and collection snapshot. Different versions, harnesses, or providers must not be compared directly; undisclosed parameters remain unknown.
For a reproduction, fix the model version, provider or client, reasoning tier, tools, task-set version, sample count, and collection date, and record failures, retries, and human corrections. Full steps are in the source notes below.
Artificial Analysis · Artificial Analysis Research Team · Original publication date 2026-02-19 · Site edit date 2026-09-20
Open original sourceGemini 3.1 Pro
Download the Tabbit client to check model access
