Skip to content
Get an API key

Live · serverless · 3 regions

Gemma 3 27B

by Google

A 27B model that reads images and text in more than 140 languages and runs on a single H100.

  • chat
  • vision
  • BF16
  • dedicated
Serverless priceUSD
Input
$0.20per 1M tokens
Output
$0.40per 1M tokens

Billed per token, no minimums.

Reference

Specifications

Model ID
gemma-3-27b
Endpoint
POST /v1/chat/completions
Publisher
Google
Tasks
chat · vision
Served context
128K tokens
Quantization
BF16
Decode speed
140 tok/smedian, one request
p50 first token
135 mslast 24 hours
Regions
us-east · us-west · eu-west
Dedicated
1× NVIDIA H100 per replica

Last 24 hours

Latency and speed

Median of each 15-minute step on the shared pool, measured at the API edge. Time to first token includes queueing; decode speed is for one request.

Time to first token · ms

135 msmedian

p95 180 ms

Decode speed · tok/s

140 tok/smedian

p5 119 tok/s

Show the data
Median of each 15-minute step over the last 24 hours, newest first
StepTime to first tokenDecode speed
15m ago202 ms109 tok/s
30m ago198 ms105 tok/s
45m ago135 ms136 tok/s
1h ago133 ms145 tok/s
1h 15m ago146 ms138 tok/s
1h 30m ago138 ms141 tok/s
1h 45m ago143 ms137 tok/s
2h ago136 ms137 tok/s
2h 15m ago135 ms142 tok/s
2h 30m ago139 ms140 tok/s
2h 45m ago139 ms138 tok/s
3h ago139 ms136 tok/s
3h 15m ago136 ms135 tok/s
3h 30m ago143 ms132 tok/s
3h 45m ago139 ms137 tok/s
4h ago138 ms136 tok/s
4h 15m ago137 ms139 tok/s
4h 30m ago134 ms137 tok/s
4h 45m ago180 ms119 tok/s
5h ago139 ms136 tok/s
5h 15m ago136 ms141 tok/s
5h 30m ago145 ms133 tok/s
5h 45m ago136 ms141 tok/s
6h ago132 ms141 tok/s
6h 15m ago142 ms135 tok/s
6h 30m ago142 ms138 tok/s
6h 45m ago136 ms135 tok/s
7h ago136 ms139 tok/s
7h 15m ago133 ms137 tok/s
7h 30m ago139 ms139 tok/s
7h 45m ago139 ms137 tok/s
8h ago139 ms137 tok/s
8h 15m ago133 ms144 tok/s
8h 30m ago140 ms140 tok/s
8h 45m ago138 ms141 tok/s
9h ago138 ms137 tok/s
9h 15m ago136 ms137 tok/s
9h 30m ago133 ms140 tok/s
9h 45m ago127 ms141 tok/s
10h ago137 ms143 tok/s
10h 15m ago127 ms144 tok/s
10h 30m ago128 ms147 tok/s
10h 45m ago133 ms144 tok/s
11h ago138 ms142 tok/s
11h 15m ago129 ms139 tok/s
11h 30m ago133 ms143 tok/s
11h 45m ago129 ms143 tok/s
12h ago124 ms141 tok/s
12h 15m ago136 ms138 tok/s
12h 30m ago130 ms140 tok/s
12h 45m ago132 ms141 tok/s
13h ago175 ms120 tok/s
13h 15m ago126 ms141 tok/s
13h 30m ago128 ms147 tok/s
13h 45m ago129 ms139 tok/s
14h ago123 ms145 tok/s
14h 15m ago130 ms141 tok/s
14h 30m ago126 ms143 tok/s
14h 45m ago156 ms126 tok/s
15h ago131 ms142 tok/s
15h 15m ago130 ms144 tok/s
15h 30m ago131 ms139 tok/s
15h 45m ago135 ms142 tok/s
16h ago132 ms141 tok/s
16h 15m ago127 ms144 tok/s
16h 30m ago128 ms147 tok/s
16h 45m ago124 ms143 tok/s
17h ago132 ms145 tok/s
17h 15m ago133 ms137 tok/s
17h 30m ago123 ms142 tok/s
17h 45m ago128 ms145 tok/s
18h ago136 ms138 tok/s
18h 15m ago133 ms141 tok/s
18h 30m ago127 ms141 tok/s
18h 45m ago127 ms146 tok/s
19h ago137 ms135 tok/s
19h 15m ago126 ms145 tok/s
19h 30m ago131 ms145 tok/s
19h 45m ago134 ms142 tok/s
20h ago131 ms140 tok/s
20h 15m ago135 ms142 tok/s
20h 30m ago137 ms141 tok/s
20h 45m ago128 ms145 tok/s
21h ago129 ms143 tok/s
21h 15m ago137 ms135 tok/s
21h 30m ago133 ms137 tok/s
21h 45m ago140 ms133 tok/s
22h ago137 ms141 tok/s
22h 15m ago133 ms137 tok/s
22h 30m ago136 ms136 tok/s
22h 45m ago196 ms106 tok/s
23h ago190 ms114 tok/s
23h 15m ago138 ms140 tok/s
23h 30m ago144 ms133 tok/s
23h 45m ago133 ms140 tok/s
24h ago143 ms140 tok/s

API

Make a request

The same request shape as every other model on /v1/chat/completions. Switching between them is a change to one string.

curl https://api.weir.example/v1/chat/completions \
  -H "Authorization: Bearer $WEIR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemma-3-27b",
    "messages": [
      { "role": "user", "content": "What does a KV cache do?" }
    ]
  }'
POST /v1/chat/completionsapi.weir.example

Dedicated

Run it on GPUs reserved for you

Replica
1× H100
Replica rate
$2.49 / hour
Peak throughput
6,000 tok/s
Break-even, 1 replica
6.06B tok/mo

Replicas reserved for you, billed by the second, with a 99.95% monthly uptime SLA. Break-even is where one replica costs the same as serverless, at half input and half output tokens.

Run on dedicated GPUsCompare costs

Compare

Similar models