openai/gpt-oss-20b pricing
gpt-oss-20b is an open-weight 21B parameter model released by OpenAI under the Apache 2.0 license. It uses a Mixture-of-Experts (MoE) architecture with 3.6B active parameters per forward pass, optimized for lower-latency inference. openai/gpt-oss-20b from DeepInfra costs $0.030 per 1M input tokens and $0.140 per 1M output tokens. Context window 131,072 tokens.
API id openai/gpt-oss-20b · DeepInfra · source · updated 2026-09-25 · general-purpose
$0.030
Input / 1M tokens
$0.140
Output / 1M tokens
—
Cached input / 1M
131K
Context window
openai/gpt-oss-20b specifications
- Provider
- DeepInfra
- API id
openai/gpt-oss-20b- Model type
- general-purpose
- Context window
- 131K tokens
- Max output
- not published
- Currency
- USD
- Price source
- Provider API
- Last checked
- 2026-09-25
OpenAI: gpt-oss-20b across providers
The same model, priced by every seller we track. Ranked by the cost of 1M input + 1M output tokens, standard tier; savings measured against buying from the vendor directly.
| Provider | Input /1M | Cached /1M | Output /1M | 1M in + 1M out | vs vendor |
|---|---|---|---|---|---|
| $0.018 | — | $0.090 | $0.11 | — | |
| $0.030 | — | $0.140 | $0.17 | — |
What openai/gpt-oss-20b costs in practice
| Workload | Input tokens | Output tokens | Estimated cost |
|---|---|---|---|
| Short chat turn | 1,000 | 500 | $0.0001 |
| Document summary | 50,000 | 2,000 | $0.0018 |
| 1M tokens in, 200K out | 1,000,000 | 200,000 | $0.06 |
Estimates use the standard tier and ignore caching discounts, so real bills with repeated prompt prefixes are typically lower.
openai/gpt-oss-20b price history
| Recorded | Input /1M | Output /1M |
|---|---|---|
| 2026-09-23 | $0.018 | $0.090 |
| 2026-08-27 | $0.030 | $0.140 |
| 2026-08-27 | $0.030 | $0.130 |
Common questions
- How much does openai/gpt-oss-20b cost per 1M tokens?
- openai/gpt-oss-20b costs $0.030 per 1M input tokens and $0.140 per 1M output tokens on the standard tier.
- What is openai/gpt-oss-20b's context window?
- openai/gpt-oss-20b accepts up to 131,072 tokens in a single request.
- Is there a cheaper DeepInfra model than openai/gpt-oss-20b?
- Yes. mistralai/Mistral-Nemo-Instruct-2407 is $0.019 per 1M input tokens versus $0.030 for openai/gpt-oss-20b, with a 131,072 token context window.
- Where does this openai/gpt-oss-20b price come from?
- From the OpenRouter catalogue, because DeepInfra does not publish machine-readable pricing. OpenRouter is a reseller, so its rate can differ from DeepInfra's own — verify before committing spend.
Head-to-head comparisons
Other DeepInfra models
- XiaomiMiMo/MiMo-V2.6-Flash $0.140
- anthropic/claude-fable-5 $10.00
- Qwen/Qwen3-VL-235B-A22B-Instruct $0.200
- Qwen/Qwen3-14B $0.120
- Qwen/Qwen3-Max-Thinking $1.20
- ByteDance/Seed-2.0-mini $0.100
- deepseek-ai/DeepSeek-V4-Flash-0731 $0.060
- zai-org/GLM-5.3-Flash $0.150
- anthropic/claude-opus-5-5 $4.00
- meta-llama/Llama-4-Scout-17B-16E-Instruct $0.100
- ibm-granite/granite-4.2-8b $0.060
- Qwen/Qwen3-30B-A3B $0.120
- openai/gpt-oss-120b-Ultra $0.200
- ByteDance/Seed-2.0-pro $0.500
- thinkingmachines/Inkling $0.950
- zai-org/GLM-4.6 $0.500