Skip to main content

Includes 694 hourly rounds up to 2026-11-16 from NIMStats by Mauro Druwel (source), measured with a different prompt and timeout. Our own probe adds a round every 30 minutes.

google

gemma-4-31b-it

Unstable

Rank 13 of 16 Last probed (22 min ago)

google/gemma-4-31b-it answered 62.8% of probes on the NVIDIA NIM free API in the last 30 days, with a median time to first token of 98.6 s and 4.2 tok/s. It ranks 13 of 16 tracked models.

62.8% Uptime 445/709 probes
43/ 100 Score 0 round wins
98.6 s First token p95 249.9 s
119.8 s Full reply median
10.2 s Best reply fastest in range
4.2tok/s Throughput median

Model vs. median

Against the median of models that answered, last 30 days.

Uptime
62.8% 77.3%
15 pts below median
First token
98.6 s 16.6 s
6.0× worse than median
Full reply
119.8 s 22.6 s
5.3× worse than median
Throughput
4.2 tok/s 18.0 tok/s
4.3× worse than median

gemma-4-31b-it median of answering models

Error breakdown

264 failed of 709 probes, last 30 days.

FailureProbesShare of failures
Timed out129 48.9%
Network error100 37.9%
Not served (404)25 9.5%
Server error (5xx)6 2.3%
Other error4 1.5%

Time to first token

Each probe. Lower is better.

0 ms125.0 s250.0 s375.0 s500.0 s10-31 14:0011-17 16:13 UTC
answered no answer

Throughput

Tokens per second after the first token, each probe. Higher is better.

0.0 tok/s12.5 tok/s25.0 tok/s37.5 tok/s50.0 tok/s10-31 14:0011-17 16:13 UTC
answered no answer

Availability heatmap

Last 7 days by UTC hour.

all answered some failed none answered not probed

Run history

Last 20 probes.

TimeResultFirst tokenFull replySpeed
Timed outn/an/an/a
Timed outn/an/an/a
OK97.7 s102.1 s33.2 tok/s
Timed outn/an/an/a
Timed outn/an/an/a
OK154.0 s160.6 s20.5 tok/s
Timed outn/an/an/a
Timed outn/an/an/a
OK59.1 s66.2 s19.3 tok/s
Timed outn/an/an/a
Timed outn/an/an/a
OK36.3 s40.6 s31.8 tok/s
OK41.7 s51.4 s9.7 tok/s
OK56.1 s65.7 s7.6 tok/s
Timed outn/an/an/a
Timed outn/an/an/a
Timed outn/an/an/a
OK249.8 s265.5 s1.9 tok/s
OK153.6 s164.8 s3.0 tok/s
OK188.2 s220.0 s2.3 tok/s

Call gemma-4-31b-it

The endpoint is OpenAI-compatible.

Create a free key at build.nvidia.com, then:

curl https://integrate.api.nvidia.com/v1/chat/completions \
  -H "Authorization: Bearer $NVIDIA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/gemma-4-31b-it",
    "messages": [{"role": "user", "content": "Hello"}],
    "max_tokens": 256,
    "stream": true
  }'

Other models answering

Top of the leaderboard right now.