Gemini 3.1 Pro Preview Matches GPT-5.4 on Intelligence Index at 20% Lower Cost
Gemini 3.1 Pro Preview is live on Google AI Studio and the Gemini API, priced at $2.00 per million input tokens and $12.00 per million output tokens — for prompts under 200K tokens. At that price, it matches GPT-5.4’s score of 57 on Artificial Analysis’s Intelligence Index v4.0, the primary quality ranking across 126 models.
GPT-5.4 costs $2.50 input and $15.00 output per million. Claude Opus 4.6 — which scores 53 on the same index, four points below — costs $5.00 input and $25.00 output. The pricing spread at the top of the quality distribution has never been wider.
The Numbers That Matter
| Model | Input / 1M | Output / 1M | AA Score | Context |
|---|---|---|---|---|
| Gemini 3.1 Flash-Lite | $0.25 | $1.50 | 34 | 1M |
| Gemini 2.5 Pro | $1.25 | $10.00 | 47 | 1M |
| Gemini 3.1 Pro Preview | $2.00 | $12.00 | 57 | 1M |
| GPT-5.4 | $2.50 | $15.00 | 57 | 1.1M |
| Claude Opus 4.6 | $5.00 | $25.00 | 53 | 1M |
Blended at a typical 25/75 input/output mix: Gemini 3.1 Pro costs $9.50 per million blended tokens versus GPT-5.4 at $11.875 and Claude Opus 4.6 at $20.00. On identical quality, Gemini 3.1 Pro is cheaper than GPT-5.4, and less than half the cost of Opus.
Pricing Cliffs
Google’s tiered pricing introduces a meaningful step function at 200K tokens. Requests exceeding 200K are billed at $4.00 input / $18.00 output — double the under-200K rate, applied to the entire request. A 201K-token prompt costs twice as much to input as a 199K-token one.
For most production workloads — customer support, coding assistance, document analysis, summarisation — prompts stay well under 200K. The cliff matters primarily for applications doing full-document or full-codebase analysis. For those use cases, the 1M context window is still the only option at this quality tier, but teams need to account for the billing step in their cost modelling.
Batch API cuts the rates in half. Context caching runs at $0.20 per million tokens. For workloads with significant prompt reuse — long system prompts, repeated reference documents — effective per-token costs drop substantially.
Preview Status Is the Real Risk
Gemini 3.1 Pro is still in preview. The old model ID (gemini-3-pro-preview) was shut down on March 9 with minimal notice, requiring developers to migrate to gemini-3.1-pro-preview. That pattern — shutting down a preview model without a long transition window — is typical for Google’s rapid iteration cadence but creates real operational risk for production systems.
The pricing is also explicitly subject to change before general availability. Google has previously adjusted Gemini pricing between preview and GA in both directions. Developers building cost-sensitive applications on the current $2.00/$12.00 rate should treat it as directionally stable rather than contractually fixed.
What This Does to the Market
Three months ago, the only models matching GPT-5.4 on quality indexes were Claude Opus 4.6 (at 2.5× the price) and GPT-5.4 itself. Gemini 3.1 Pro’s price-performance point breaks that pairing and forces a genuine three-way cost comparison at the top of the quality distribution for the first time.
The competitive pressure on OpenAI is direct: GPT-5.4 now needs to justify a 25% input and 25% output price premium over an equivalent-quality model. The value case is primarily developer ecosystem lock-in, context window advantage (1.1M versus 1M), and the OpenAI brand’s enterprise trust. Those advantages are real but quantifiable — and at $2.50 versus $2.00 input, they are no longer structurally unchallenged.