DeepSeek V3.2 vs Gemma 4 31B
API pricing, context, and independently measured performance, side by side. Gemma 4 31B starts 57% cheaper on input than DeepSeek V3.2 at list prices.
DeepSeek V3.2
DeepSeek · released Dec 1, 2025
Best input$0.209 / MTok
Best output$0.31 / MTok
Context164K
Providers15
Fastest measured0.7s° TTFT · 102 tok/s
Open weightsToolsReasoningStructured outputPrompt caching
Gemma 4 31B
Google · released Apr 2, 2026
Best input$0.09 / MTok
Best output$0.34 / MTok
Context262K
Providers16
Fastest measured0.31s° TTFT · 205 tok/s
Open weightsVisionToolsReasoningStructured outputPrompt caching
Cheapest providers, measuredUSD / MTOK
| MODEL | PROVIDER | INPUT | OUTPUT | TTFT | TPS | SMOKE |
|---|---|---|---|---|---|---|
| DeepSeek V3.2 | GMIGMICloud | $0.209 | $0.31 | 4.1s° | 33° tok/s | 5/5 |
| DeepSeek V3.2 | STStreamLake | $0.214 | $0.322 | 1.8s° | 28° tok/s | 5/5 |
| DeepSeek V3.2 | DIDigitalOcean | $0.25 | $0.80 | 0.7s° | 120° tok/s | 5/5 |
| Gemma 4 31B |  DeepInfra | $0.09 | $0.34 | 0.84s° | 56° tok/s | 5/5 |
| Gemma 4 31B | CWCoreWeave | $0.10 | $0.34 | 1.6s° | 38° tok/s | 5/5 |
| Gemma 4 31B |  Venice | $0.12 | $0.36 | 1.5s° | 81° tok/s | 5/5 |
TTFT/TPS: ModelIndex-measured medians (P50, trailing 21d) · chat10k profile (~10K in / 200 out) · ° = measured via OpenRouter routing (adds a hop) · SMOKE: of 5 programmatic capability probes passed · methodology v1.1 — full details on each model page