gpt-oss-20b pricing across providers
Across tracked provider offerings, input runs $0.018–$0.075 per million tokens and output $0.090–$0.300 per million tokens. The cheapest tracked provider offering for a typical agentic mix (2.5% input, 97% cached input, 0.5% output) is Darkbloom at $0.0096 per million tokens.
Prices are USD per million tokens and reflect each provider offering — they are not a single intrinsic model price. 6 of 10 offerings publish a cache-read rate; 7 of 10 publish a 30-minute endpoint uptime figure.
Open the calculator filtered to gpt-oss-20b
Is this the right model for your workload? Compare open-weight picks and providers for your use case →
Daily price history
Loading price history…
Daily snapshots are retained for up to 90 days; history begins with the first recorded snapshot.
Provider offerings for gpt-oss-20b
Ranked by effective cost at a typical agentic mix. Quantized and tier variants stay separate rows.
| Provider | Quant | Input $/M | Output $/M | Cache $/M | Blended $/M | Uptime (30m) |
|---|---|---|---|---|---|---|
| Darkbloom | fp8 | $0.018 | $0.090 | $0.0090 | $0.0096 | 100.00% |
| Consensusprotocol | — | $0.040 | $0.190 | $0.010 | $0.012 | — |
| Akashml | fp4 | $0.020 | $0.100 | — | $0.020 | 100.00% |
| Parasail | fp4 | $0.030 | $0.150 | $0.020 | $0.021 | 100.00% |
| Dekallm | bf16 | $0.029 | $0.140 | $0.029 | $0.030 | 100.00% |
| Coreweave | fp4 | $0.030 | $0.130 | $0.030 | $0.030 | 100.00% |
| DeepInfra | — | $0.030 | $0.140 | — | $0.031 | — |
| Groq | unknown | $0.075 | $0.300 | $0.037 | $0.040 | 100.00% |
| SiliconFlow | fp8 | $0.040 | $0.180 | — | $0.041 | 66.67% |
| Amazon | unknown | $0.070 | $0.150 | — | $0.070 | — |
Provider-offering pricing refreshed 2026-10-29. Uptime is the 30-minute endpoint metric where a provider publishes it. Verify rates on the provider's official pricing page before committing spend.