Gemini 3

Why this family matters

Gemini 3 is Google's current generation, structured as a three-tier ladder (Pro, Flash, Flash Lite). Two facts pull most teams onto this family. First, the per-token pricing is aggressive against the closed-flagship field: Gemini 3 Flash lists at $0.5 input / $3 output per million tokens and Gemini 3.1 Flash Lite at $0.25 / $1.5, which sits below the equivalent tiers on most competitor pricing tables we track. Second, Gemini 3 Pro 3.1 lands at Quality Score 104.3 (#5 of 186 models in our index), which puts the flagship inside the same cluster as the closed-frontier tier, not below it.

The combination is unusual. Most families are either cheap-and-mid or expensive-and-frontier; Gemini 3 ships both ends of the ladder simultaneously, which makes "which tier" the entire decision.

Which variant to start with

Default to google-gemini-3-flash. At Quality Score 88.9 (#32 of 186) and $0.5 input / $3 output, it is the family's price-quality sweet spot for chat, summarization, and tool-augmented assistants. For most teams shipping API-backed product features, this is the practical default.

Step up to google-gemini-3-pro when the workload visibly benefits from the additional headroom. The 3.1 variant lands at QS 104.3 against Flash's 88.9, which is a meaningful jump on the hardest reasoning, coding, and multi-step planning evals (GPQA Diamond 94.3, LiveBench 79.93). The price gap is roughly 4x on input ($2 vs $0.5) and 4x on output ($12 vs $3), so the workload needs to be one where the score delta translates to a measurable product win.

Drop to google-gemini-3-1-flash-lite for high-volume chat at scale. Quality Score 81.1 vs Flash's 88.9 is a real gap, but at

$0.25 /

the per-token saving compounds quickly on workloads dominated by repetitive low-stakes turns.

When to deviate:

• Long-context RAG: use google-gemini-3-pro. The 3.0 variant in our index ships with a 1M-token context window at the same headline pricing. Reach for it when document scale and faithful retrieval over long inputs dominate the workload.
• Hardest-tier reasoning workloads: Gemini 3 Pro 3.1 is competitive at the very top of our reasoning leaderboards (GPQA Diamond rank 3, LiveBench rank 4). Run a side-by-side against your closed-flagship alternative on the specific reasoning benchmark that matters; on the numbers we have, the cost-quality trade-off is genuinely close.
• You already use Flash for everything: before adding Pro to the rotation, run an A/B on your specific eval. The score gap on benchmarks is real; whether it shows up on your traffic distribution is the question worth answering with data, not vibes.

Where the data is weak

We aggregate benchmark scores from multiple sources but coverage is uneven across this family. Specifically:

• Gemini 3 Pro has two minor versions in our index (3.0 and 3.1) with substantially different scores. Pro 3.0 sits at QS 95.0; Pro 3.1 at QS 104.3. When the article quotes a number, it is for the specific minor version named; do not collapse the line to a single Pro score.
• Context windows are partially declared. Our index lists the 1M-token window only on Gemini 3 Pro 3.0; Pro 3.1, Flash, and Flash Lite show the field as unset. That is a coverage gap, not evidence of a smaller window. Verify the limit on the deployment surface you actually use before committing on a long-document workload.
• Flash Lite has thinner benchmark coverage than the other tiers. Several benchmarks (SWE-Bench Verified, AIME 2025) carry numbers for Flash and Pro but not Flash Lite at last verification. Treat its scores as directional outside the headline benchmarks (Quality Score, Arena ELO, LiveBench, GPQA Diamond).
• Pricing on this page is the published API list price. Vertex AI routing, batch pricing, and enterprise agreements can change the unit economics. List price is a calibration anchor, not the cost ceiling.

If you are making a procurement decision, the variant table on this page is the load-bearing artifact. Cross-check pricing against Google's own docs (and Vertex AI routing if you go through that path) before you commit.

When to reach for which alternative

• Open-weights deployment is a requirement: Gemini is API-only. The conversation moves to open-weights families (Qwen3, DeepSeek). On the cost-per-quality axis at the chat workhorse tier, DeepSeek V4 Flash ($0.098 / $0.197 with QS 78.1) is the price anchor to beat.
• Closed-flagship reasoning at the absolute top end: Claude Opus 4.7-thinking lands at QS 107.8 in our index, slightly above Gemini 3 Pro 3.1 (QS 104.3). On any specific benchmark the ranking can flip; compare on the workload that matters before treating the headline QS difference as decisive.
• Previous-generation Gemini is already in production: the Gemini 2 line is on the sibling gemini-2 surface in our index. For some workloads the migration cost to 3 may not be earned by the score delta, particularly if Vertex routing or fine-tunes are tied to the older line.

Sources worth reading

• Google AI Studio pricing: authoritative price list for the Gemini API
• Gemini API model docs: variant identifiers, context windows, modality coverage
• Google AI for Developers blog: release notes for new generations and pricing changes