by datastudy.nl

Head-to-head comparisons of frontier AI models, rebuilt daily from public benchmark data

Google

Gemini 3.8 Flash

Released Sep 2026 · knowledge cutoff Mar 2026. Proprietary, available through the provider's API.

Gemini 3.8 Flash is built by Google and has 13 benchmark results tracked in the public leaderboard. It supports multimodal input, with a context window of 1.05M tokens. At the blended 8:1 input-to-output rate, it costs $1.08 per million tokens and delivers about 86.7 tokens per second.

Blended price$1.08per 1M tokens (8:1)
Context1.05Mtokens
Output speed86.7tokens / sec
Benchmarks13results tracked

Where Gemini 3.8 Flash ranks

The chart below ranks every model in the comparison by overall capability, which is the average of a model's scores across the eight workload axes. Gemini 3.8 Flash is the red bar. This is the single number to check first: it tells you where the model sits in the current pecking order before you dig into the detail.

Bar chart ranking all ten models by overall capability index, with Gemini 3.8 Flash highlighted in red.
Every model ranked by overall capability index (0 to 100), averaged across eight workload axes. Gemini 3.8 Flash is the red bar. Data Today, from llm-stats.com public benchmarks.

Workload strengths

Each row below is the average normalized score across every benchmark tagged with that workload category. A higher number means the model performs better on that kind of task relative to the maximum possible score on those benchmarks. Categories where the model has no published results are shown as n/a.

WorkloadIndex (0–1)
Reasoning58%
Coding61%
Agents55%
Mathn/a
Vision67%
Long Context87%
Writingn/a
Knowledge45%

How Gemini 3.8 Flash stacks up against every other model

Below is every head-to-head comparison involving Gemini 3.8 Flash. Each card shows the win/loss tally across all workload axes and headline benchmarks. Tap any matchup for the full breakdown with per-benchmark scores, pricing, and workload recommendations.


Choose a matchup above, or start from the hub to pick any two models.