Google: Gemma 4 31B
google/gemma-4-31b-itGemma 4 31B Instruct is Google DeepMind's 30.7B dense multimodal model supporting text and image input with text output.
Modalities
In / out per 1M$0.20 / $0.40
Context256K
ReleasedApr 2, 2026
Providers
15 providers serve this model upstream.
| Provider | Input /M | Output /M | Cache read /M | Latency P50 | Throughput P50 | Uptime |
|---|---|---|---|---|---|---|
 CoreWeavefp4 | $0.10 | $0.34 | $0.10 | 0.43s | 41 tps | 100% |
 ModelRun [by Modular]fp4 | $0.75 | $1.00 | $0.20 | 0.09s | 183 tps | 100% |
 DeepInfra (fp4)us · fp4 | $0.14 | $0.51 | $0.07 | 0.56s | 33 tps | 99.97% |
| $0.14 | $0.40 | Variable | 0.93s | 75 tps | 99.97% | |
| $0.15 | $0.40 | $0.06 | 1.57s | 22 tps | 99.86% | |
| DeepInfra (fp8)us · fp8 | $0.20 | $0.40 | Variable | 0.75s | 24 tps | 99.8% |
| DeepInfra (fp8)us · fp8 | $0.16 | $0.32 | Variable | — | — | 99.79% |
| $0.12 | $0.36 | $0.09 | 0.64s | 33 tps | 99.77% | |
| DeepInfra (fp8)us · fp8 | $0.30 | $0.60 | Variable | 0.51s | 21 tps | 99.57% |
| $0.14 | $0.40 | $0.14 | 0.53s | 40 tps | 98.41% | |
| $0.36 | $1.09 | $0.18 | 0.89s | 26 tps | 97.05% | |
| $0.38 | $1.15 | Variable | 1.90s | 148 tps | 96.34% | |
 SiliconFlowfp8 | $0.75 | $1.00 | $0.25 | 2.36s | 9 tps | 94.91% |
| $0.12 | $0.37 | $0.01 | 1.84s | 23 tps | 89.34% | |
| $0.14 | $0.40 | Variable | 0.80s | 23 tps | 71.14% |
Market-wide measurements, refreshed every 15 minutes. They describe who serves Gemma 4 31B in general, not how ProjectCOZY routes your requests.
Pricing
List rates per 1M tokens — what Gemma 4 31B costs pay-as-you-go elsewhere, not what you pay here.
Input
$0.20
per 1M tokens
Output
$0.40
per 1M tokens
Cache read
—
not published
Included in every pass
Gemma 4 31B is included in every ProjectCOZY pass at no extra cost — one flat rate, no per-token metering. The rates above are there for comparison.
Performance
Time to first token and output speed across every provider, over the last 3 days.
0.0s20.0s40.0s60.0s80.0s
Oct 16, 12 AMOct 16, 8 AMOct 16, 4 PMOct 17, 12 AMOct 17, 8 AMOct 17, 4 PMOct 18, 12 AMOct 18, 8 AMOct 18, 4 PMOct 19, 12 AMOct 19, 8 AM
SambaNova Crusoe Friendli DeepInfra (fp4) (us) SiliconFlow Chutes Venice io.net CoreWeave NovitaAI Parasail ModelRun [by Modular] DeepInfra (fp8) (us) DeepInfra (fp8) (us) DeepInfra (fp8) (us)
0 tok/s250 tok/s500 tok/s750 tok/s1000 tok/s
Oct 16, 12 AMOct 16, 8 AMOct 16, 4 PMOct 17, 12 AMOct 17, 8 AMOct 17, 4 PMOct 18, 12 AMOct 18, 8 AMOct 18, 4 PMOct 19, 12 AMOct 19, 8 AM
SambaNova Crusoe Friendli DeepInfra (fp4) (us) SiliconFlow Chutes Venice io.net CoreWeave NovitaAI Parasail ModelRun [by Modular] DeepInfra (fp8) (us) DeepInfra (fp8) (us) DeepInfra (fp8) (us)
Uptime
100% of requests were served successfully over the last 3 days, across all providers.
Last 3 daysAvailability 99.92%
Oct 16, 10:00 AMNow
Availability over the last 24 hoursAvailability 99.9%
100% 98% 97% 95%
Oct 18, 10:20 AMNow
Apps
Tools that most often run Gemma 4 31B, by tokens this week.
Activity
1986.5B tokens processed market-wide in the last month.
025.0B50.0B75.0B100.0B
Sep 19Sep 24Sep 29Oct 4Oct 9Oct 14Oct 19
Tokens
FAQ
Which model id do I use for Gemma 4 31B?
google/gemma-4-31b-it. Send it as the model field from any OpenAI-compatible client, against http://192.168.1.55/v1.
How much context does Gemma 4 31B hold?
262,144 tokens per request, of which up to 16,384 can be the response.
What can it accept and return?
Input: image, text, video. Output: text.
What does Gemma 4 31B cost on a pass?
Nothing beyond the pass itself. Pay-as-you-go elsewhere it lists at $0.20 per 1M input tokens and $0.40 per 1M output tokens.
Who built Gemma 4 31B?
Google, in the Gemma family.
Explore
Other models your pass can call.
