GPT-OSS 20B API pricing

14 providers serve GPT-OSS 20B. GPT-OSS 20B is an open-weight model (Apache-2.0) with 21B total parameters, 3.6B active per token (mixture-of-experts), up to 131K context. Prices below are per 1M tokens, cheapest input first (self-hosting math further down).

Refreshed 2026-08-23 · list prices from provider APIs
Input floor $0.015/1M at Darkbloom Output floor $0.070/1M at Darkbloom Find your workload winner →
Provider Input $/1M Output $/1M Cache read $/1M Context
Darkbloom $0.015 $0.070output floor - 131K
OpenRouter $0.020 $0.100 - 131K Use
DeepInfra $0.040 $0.150 - 131K Use
OVHcloud $0.040 $0.150 - 131K Use
Novita AI $0.040 $0.150 - 131K Use
Together AI $0.050 $0.200 - 128K Use
Databricks $0.070 $0.300 - 131K Use
Fireworks AI $0.070 $0.300 $0.035 131K Use
Tensormesh $0.070 $0.280 - 131K
Groq $0.075 $0.300 $0.037 131K Use
Google Vertex AI $0.075 $0.300 - 131K Use
AWS Bedrock $0.075 $0.300 - 131K Use
Replicate $0.090 $0.360 - - Use
Cloudflare $0.200 $0.300 - 128K Use

The cheapest way to run GPT-OSS 20B

On an illustrative 70M-input / 30M-output monthly workload, today's lowest listed cost is roughly $3.12/month, at Darkbloom. Input-heavy, output-heavy and cache-heavy workloads can produce different winners. Use the live mix calculator below rather than combining floors from two different providers. On output-token cost alone, a busy self-hosted deployment can work out cheaper (~$0.0065/1M output on V100). See the math below. Run your own numbers in the breakeven calculator, or read the breakeven math.

Your actual monthly cost

Uses today's list prices

Cache read uses the listed cache rate where available; otherwise it falls back to normal input price. Batch, write-cache, volume and negotiated discounts are excluded.

Or run it yourself: GPT-OSS 20B self-hosting economics

GPT-OSS 20B is open-weight (Apache-2.0), so the API price above competes with the GPU-hour market. 21B total parameters (MoE, ~3.6B active per token) needs roughly 29 GB VRAM at FP8 or 15 GB at INT4, KV-cache headroom included.

GPU that fits (FP8, single card)VRAMFrom $/hrCheapest at
V100 32 GB $0.022marketplace Vast.ai Rent
A6000 48 GB $0.287marketplace Vast.ai Rent
RTX 5090 32 GB $0.321marketplace Vast.ai Rent
RTX PRO 4500 Blackwell 32 GB $0.340community RunPod Rent
A40 48 GB $0.350community RunPod Rent
6000 Ada 48 GB $0.388marketplace Vast.ai Rent
Breakeven estimate: a well-batched vLLM deployment on one V100 ($0.022/hr) at ~1875 aggregate tok/s and 50% utilization works out to roughly $0.0065 per 1M output tokens, versus $0.070 via the cheapest output-token API (Darkbloom). Self-hosting wins on cost if you can keep the GPU busy. Planning estimate: throughput varies with hardware, quantization, batch size and context. Tune it in the calculator.

Related models

ModelCheapest in $/1MCheapest out $/1MProviders
GPT-OSS 120B $0.050 $0.250 20
GPT-5 $1.25 $10.00 8
GPT-5 Mini $0.250 $1.20 7
GPT-5 Nano $0.050 $0.400 7
GPT-4o $2.50 $10.00 6
GPT-4o-Mini $0.150 $0.600 6

FAQ

What is the cheapest API for GPT-OSS 20B?

As of 2026-08-23, the lowest input price for GPT-OSS 20B is Darkbloom at $0.015 per 1M input tokens. The lowest output price is Darkbloom at $0.070 per 1M output tokens. The cheapest provider for a real workload depends on its input/output mix. The next-lowest input price is OpenRouter at $0.020, a 38% difference.

How much VRAM do you need to self-host GPT-OSS 20B?

GPT-OSS 20B has 21B parameters, so plan for roughly 29 GB of VRAM at FP8 or 15 GB at INT4/AWQ, KV-cache headroom included. A single V100 (32 GB) fits it.

← All models · GPU rental prices · Breakeven calculator · Get this data as JSON