Skip to content
Get an API key

Live · serverless · 3 regions

Qwen3 32B

by Alibaba

A dense 32B model that can think step by step or answer directly, and fits on a single H100.

  • chat
  • reasoning
  • FP8
  • dedicated
Serverless priceUSD
Input
$0.30per 1M tokens
Output
$0.50per 1M tokens

Billed per token, no minimums.

Reference

Specifications

Model ID
qwen3-32b
Endpoint
POST /v1/chat/completions
Publisher
Alibaba
Tasks
chat · reasoning
Served context
128K tokens
Quantization
FP8
Decode speed
130 tok/smedian, one request
p50 first token
150 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

150 msmedian

p95 190 ms

Decode speed · tok/s

130 tok/smedian

p5 112 tok/s

Show the data
Median of each 15-minute step over the last 24 hours, newest first
StepTime to first tokenDecode speed
15m ago157 ms128 tok/s
30m ago147 ms130 tok/s
45m ago150 ms127 tok/s
1h ago151 ms131 tok/s
1h 15m ago161 ms124 tok/s
1h 30m ago157 ms123 tok/s
1h 45m ago152 ms131 tok/s
2h ago152 ms128 tok/s
2h 15m ago151 ms133 tok/s
2h 30m ago157 ms125 tok/s
2h 45m ago149 ms131 tok/s
3h ago152 ms129 tok/s
3h 15m ago158 ms125 tok/s
3h 30m ago156 ms128 tok/s
3h 45m ago158 ms128 tok/s
4h ago157 ms125 tok/s
4h 15m ago160 ms124 tok/s
4h 30m ago156 ms124 tok/s
4h 45m ago161 ms126 tok/s
5h ago211 ms105 tok/s
5h 15m ago157 ms130 tok/s
5h 30m ago159 ms125 tok/s
5h 45m ago151 ms128 tok/s
6h ago156 ms126 tok/s
6h 15m ago151 ms132 tok/s
6h 30m ago153 ms130 tok/s
6h 45m ago158 ms126 tok/s
7h ago157 ms129 tok/s
7h 15m ago153 ms131 tok/s
7h 30m ago157 ms130 tok/s
7h 45m ago153 ms127 tok/s
8h ago154 ms130 tok/s
8h 15m ago214 ms98 tok/s
8h 30m ago155 ms126 tok/s
8h 45m ago156 ms124 tok/s
9h ago144 ms132 tok/s
9h 15m ago142 ms136 tok/s
9h 30m ago144 ms129 tok/s
9h 45m ago145 ms136 tok/s
10h ago150 ms127 tok/s
10h 15m ago153 ms132 tok/s
10h 30m ago152 ms130 tok/s
10h 45m ago138 ms135 tok/s
11h ago150 ms133 tok/s
11h 15m ago150 ms129 tok/s
11h 30m ago138 ms135 tok/s
11h 45m ago139 ms138 tok/s
12h ago142 ms135 tok/s
12h 15m ago138 ms133 tok/s
12h 30m ago139 ms136 tok/s
12h 45m ago144 ms134 tok/s
13h ago148 ms135 tok/s
13h 15m ago141 ms132 tok/s
13h 30m ago141 ms137 tok/s
13h 45m ago139 ms137 tok/s
14h ago149 ms132 tok/s
14h 15m ago149 ms127 tok/s
14h 30m ago144 ms134 tok/s
14h 45m ago190 ms112 tok/s
15h ago147 ms133 tok/s
15h 15m ago138 ms139 tok/s
15h 30m ago137 ms134 tok/s
15h 45m ago146 ms130 tok/s
16h ago144 ms134 tok/s
16h 15m ago138 ms136 tok/s
16h 30m ago141 ms135 tok/s
16h 45m ago136 ms137 tok/s
17h ago149 ms130 tok/s
17h 15m ago149 ms130 tok/s
17h 30m ago143 ms131 tok/s
17h 45m ago141 ms132 tok/s
18h ago144 ms129 tok/s
18h 15m ago138 ms134 tok/s
18h 30m ago150 ms130 tok/s
18h 45m ago149 ms127 tok/s
19h ago139 ms137 tok/s
19h 15m ago151 ms130 tok/s
19h 30m ago139 ms137 tok/s
19h 45m ago146 ms133 tok/s
20h ago149 ms129 tok/s
20h 15m ago146 ms133 tok/s
20h 30m ago149 ms133 tok/s
20h 45m ago143 ms136 tok/s
21h ago143 ms130 tok/s
21h 15m ago150 ms132 tok/s
21h 30m ago155 ms132 tok/s
21h 45m ago157 ms130 tok/s
22h ago193 ms112 tok/s
22h 15m ago154 ms128 tok/s
22h 30m ago151 ms129 tok/s
22h 45m ago147 ms133 tok/s
23h ago154 ms126 tok/s
23h 15m ago157 ms129 tok/s
23h 30m ago202 ms106 tok/s
23h 45m ago158 ms125 tok/s
24h ago153 ms128 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": "qwen3-32b",
    "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
5,000 tok/s
Break-even, 1 replica
4.54B 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