Llama 3.3 70B Instruct: providers compared
Every endpoint serving this model, side by side. ◆ marks the best value in each column — per model, never fleet-wide.
Current state
| State | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|
Groqgroq | Healthy | 100%◆ (best) | 99.86% | 99.82%◆ (best) | 99.78% | 383ms◆ (best) | 1.38s◆ (best) | 253 tps◆ (best) | 131K | $0.590 / $0.790 |
| Healthy | 100%◆ (best) | 97.17% | 92.31% | 97.32% | 1.30s | 4.27s | 37 tps | 12K | $0.135 / $0.400 | |
| Healthy | 99.68% | 99.49% | 98.92% | 99.34% | 1.64s | 5.88s | 56 tps | 131K | $0.220 / $0.500 | |
| Healthy | 99.61% | 99.38% | 98.31% | 98.02% | 764ms | 21.3s | 9.0 tps | 131K | $0.100 / $0.320◆ (best) | |
| Healthy | 98.90% | 99.03% | 98.11% | 97.76% | 482ms | 5.85s | 36 tps | 24K | $0.293 / $2.25 | |
| Healthy | 98.62% | 99.62% | 99.53% | 99.57% | 2.28s | 12.4s | 26 tps | 131K | $0.200 / $0.520 | |
| Healthy | 97.87% | 99.52% | 99.73% | 99.83% | 1.44s | 11.3s | 53 tps | 128K | $0.710 / $0.710 | |
SambaNovasambanova-turbo | Healthy | 95.41% | 98.48% | 99.11% | 98.84% | 956ms | 4.83s | 128 tps | 131K | $0.450 / $0.900 |
Togethertogether | Down | 90.88% | 92.61% | 98.58% | 97.97% | 3.60s | 16.7s | 26 tps | 131K | $1.04 / $1.04 |
| No data | — | — | — | 99.90% | — | — | — | 131K | $0.250 / $0.750 | |
Googlegoogle-vertex | No data | — | — | — | 99.13% | — | — | — | 128K | $0.720 / $0.720 |
Googlegoogle-vertex/us-central1 | No data | — | — | — | 99.98%◆ (best) | — | — | — | 128K | $0.720 / $0.720 |
| No data | — | — | — | 53.83% | — | — | — | 131K | $0.130 / $0.400 | |
PrimeIntellectprimeintellect | No data | — | — | 99.65% | 99.65% | — | — | — | 131K | $1.04 / $2.25 |
- Groq100%◆ (best)Healthy24h 99.86%
- 7d
- 99.82%◆ (best)
- 30d
- 99.78%
- p50
- 383ms◆ (best)
- p99
- 1.38s◆ (best)
- Throughput
- 253 tps◆ (best)
- Context
- 131K
- $/Mtok
- $0.590 / $0.790
groq - Novita100%◆ (best)Healthy24h 97.17%
- 7d
- 92.31%
- 30d
- 97.32%
- p50
- 1.30s
- p99
- 4.27s
- Throughput
- 37 tps
- Context
- 12K
- $/Mtok
- $0.135 / $0.400
novita/bf16bf16 - Parasail99.68%Healthy24h 99.49%
- 7d
- 98.92%
- 30d
- 99.34%
- p50
- 1.64s
- p99
- 5.88s
- Throughput
- 56 tps
- Context
- 131K
- $/Mtok
- $0.220 / $0.500
parasail/fp8fp8 - DeepInfra99.61%Healthy24h 99.38%
- 7d
- 98.31%
- 30d
- 98.02%
- p50
- 764ms
- p99
- 21.3s
- Throughput
- 9.0 tps
- Context
- 131K
- $/Mtok
- $0.100 / $0.320◆ (best)
deepinfra/turbofp8 - Cloudflare98.90%Healthy24h 99.03%
- 7d
- 98.11%
- 30d
- 97.76%
- p50
- 482ms
- p99
- 5.85s
- Throughput
- 36 tps
- Context
- 24K
- $/Mtok
- $0.293 / $2.25
cloudflare/fp8fp8 - AkashML98.62%Healthy24h 99.62%
- 7d
- 99.53%
- 30d
- 99.57%
- p50
- 2.28s
- p99
- 12.4s
- Throughput
- 26 tps
- Context
- 131K
- $/Mtok
- $0.200 / $0.520
akashml/fp8fp8 - CoreWeave97.87%Healthy24h 99.52%
- 7d
- 99.73%
- 30d
- 99.83%
- p50
- 1.44s
- p99
- 11.3s
- Throughput
- 53 tps
- Context
- 128K
- $/Mtok
- $0.710 / $0.710
coreweave/fp16fp16 - SambaNova95.41%Healthy24h 98.48%
- 7d
- 99.11%
- 30d
- 98.84%
- p50
- 956ms
- p99
- 4.83s
- Throughput
- 128 tps
- Context
- 131K
- $/Mtok
- $0.450 / $0.900
sambanova-turbo - Together90.88%Down24h 92.61%
- 7d
- 98.58%
- 30d
- 97.97%
- p50
- 3.60s
- p99
- 16.7s
- Throughput
- 26 tps
- Context
- 131K
- $/Mtok
- $1.04 / $1.04
together - No data24h —
- 7d
- —
- 30d
- 99.90%
- p50
- —
- p99
- —
- Throughput
- —
- Context
- 131K
- $/Mtok
- $0.250 / $0.750
crusoe/bf16bf16 - No data24h —
- 7d
- —
- 30d
- 99.13%
- p50
- —
- p99
- —
- Throughput
- —
- Context
- 128K
- $/Mtok
- $0.720 / $0.720
google-vertex - No data24h —
- 7d
- —
- 30d
- 99.98%◆ (best)
- p50
- —
- p99
- —
- Throughput
- —
- Context
- 128K
- $/Mtok
- $0.720 / $0.720
google-vertex/us-central1 - No data24h —
- 7d
- —
- 30d
- 53.83%
- p50
- —
- p99
- —
- Throughput
- —
- Context
- 131K
- $/Mtok
- $0.130 / $0.400
nebius/fp8fp8 - No data24h —
- 7d
- 99.65%
- 30d
- 99.65%
- p50
- —
- p99
- —
- Throughput
- —
- Context
- 131K
- $/Mtok
- $1.04 / $2.25
primeintellect
Measured 1m ago. Latency and throughput are OpenRouter's rolling 30-minute windows. ◆ marks the best value in each column among these endpoints.
Price vs latency
One dot per endpoint at the latest round · further left = faster, lower = cheaper · hover a dot for its tag
History
Availability
Axis starts at 90% so small differences stay visible; it expands to 0 when an endpoint falls below.
Time to first token
Logarithmic axis — the fleet spans two orders of magnitude. P50 is OpenRouter's rolling 30-minute window, averaged over the hour.
Throughput
Median tokens per second.