Skip to content
Get an API key

Live · serverless · 3 regions

DeepSeek R1 Distill Llama 70B

by DeepSeek

R1's reasoning style distilled into a 70B Llama, with shorter thinking and a lower price than the full model.

  • reasoning
  • FP8
  • dedicated
Serverless priceUSD
Input
$0.58per 1M tokens
Output
$0.78per 1M tokens

Billed per token, no minimums.

Reference

Specifications

Model ID
deepseek-r1-distill-llama-70b
Endpoint
POST /v1/chat/completions
Publisher
DeepSeek
Tasks
reasoning
Served context
128K tokens
Quantization
FP8
Decode speed
90 tok/smedian, one request
p50 first token
230 mslast 24 hours
Regions
us-east · us-west · eu-west
Dedicated
2× 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

230 msmedian

p95 246 ms

Decode speed · tok/s

90 tok/smedian

p5 85 tok/s

Show the data
Median of each 15-minute step over the last 24 hours, newest first
StepTime to first tokenDecode speed
15m ago229 ms93 tok/s
30m ago245 ms86 tok/s
45m ago246 ms85 tok/s
1h ago243 ms88 tok/s
1h 15m ago246 ms85 tok/s
1h 30m ago235 ms90 tok/s
1h 45m ago237 ms88 tok/s
2h ago240 ms88 tok/s
2h 15m ago232 ms92 tok/s
2h 30m ago227 ms92 tok/s
2h 45m ago230 ms91 tok/s
3h ago242 ms86 tok/s
3h 15m ago236 ms88 tok/s
3h 30m ago231 ms90 tok/s
3h 45m ago231 ms90 tok/s
4h ago227 ms88 tok/s
4h 15m ago242 ms86 tok/s
4h 30m ago295 ms76 tok/s
4h 45m ago247 ms86 tok/s
5h ago235 ms89 tok/s
5h 15m ago245 ms89 tok/s
5h 30m ago238 ms87 tok/s
5h 45m ago230 ms92 tok/s
6h ago235 ms88 tok/s
6h 15m ago225 ms92 tok/s
6h 30m ago240 ms86 tok/s
6h 45m ago234 ms92 tok/s
7h ago222 ms91 tok/s
7h 15m ago230 ms88 tok/s
7h 30m ago223 ms90 tok/s
7h 45m ago230 ms91 tok/s
8h ago241 ms86 tok/s
8h 15m ago225 ms90 tok/s
8h 30m ago237 ms89 tok/s
8h 45m ago219 ms92 tok/s
9h ago221 ms93 tok/s
9h 15m ago239 ms89 tok/s
9h 30m ago232 ms92 tok/s
9h 45m ago297 ms77 tok/s
10h ago223 ms89 tok/s
10h 15m ago220 ms94 tok/s
10h 30m ago234 ms91 tok/s
10h 45m ago231 ms90 tok/s
11h ago215 ms93 tok/s
11h 15m ago228 ms91 tok/s
11h 30m ago214 ms92 tok/s
11h 45m ago232 ms89 tok/s
12h ago222 ms91 tok/s
12h 15m ago229 ms90 tok/s
12h 30m ago226 ms94 tok/s
12h 45m ago225 ms89 tok/s
13h ago227 ms93 tok/s
13h 15m ago279 ms79 tok/s
13h 30m ago229 ms92 tok/s
13h 45m ago231 ms89 tok/s
14h ago223 ms93 tok/s
14h 15m ago221 ms92 tok/s
14h 30m ago230 ms91 tok/s
14h 45m ago220 ms93 tok/s
15h ago220 ms91 tok/s
15h 15m ago212 ms96 tok/s
15h 30m ago224 ms93 tok/s
15h 45m ago231 ms93 tok/s
16h ago222 ms92 tok/s
16h 15m ago230 ms88 tok/s
16h 30m ago211 ms93 tok/s
16h 45m ago209 ms94 tok/s
17h ago229 ms88 tok/s
17h 15m ago219 ms95 tok/s
17h 30m ago229 ms91 tok/s
17h 45m ago226 ms89 tok/s
18h ago233 ms91 tok/s
18h 15m ago226 ms92 tok/s
18h 30m ago234 ms88 tok/s
18h 45m ago212 ms96 tok/s
19h ago228 ms93 tok/s
19h 15m ago220 ms93 tok/s
19h 30m ago224 ms94 tok/s
19h 45m ago229 ms89 tok/s
20h ago233 ms88 tok/s
20h 15m ago224 ms93 tok/s
20h 30m ago237 ms88 tok/s
20h 45m ago239 ms87 tok/s
21h ago235 ms90 tok/s
21h 15m ago224 ms93 tok/s
21h 30m ago239 ms90 tok/s
21h 45m ago232 ms89 tok/s
22h ago223 ms89 tok/s
22h 15m ago221 ms92 tok/s
22h 30m ago240 ms88 tok/s
22h 45m ago243 ms87 tok/s
23h ago231 ms91 tok/s
23h 15m ago241 ms86 tok/s
23h 30m ago245 ms89 tok/s
23h 45m ago235 ms90 tok/s
24h ago245 ms90 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": "deepseek-r1-distill-llama-70b",
    "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
2× H100
Replica rate
$4.98 / hour
Peak throughput
5,800 tok/s
Break-even, 1 replica
5.35B 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