gpt-oss-20b API pricing
OpenAI · Budget price tier · Apache 2.0 · released 2025-08. Prices in USD per 1 million tokens. Data snapshot: 2026-10-22.
gpt-oss-20b has held the same list price since 2026-10-18 and is the 6th cheapest of 164 models in the Budget price tier. None of its 10 hosts undercuts the price we quote. It is rated a stable release but does not meet all four criteria of the FinOps Friendly badge.
- Input: $0.018 per 1M tokens
- Output: $0.09 per 1M tokens
- Batch API (in/out): Not published
- Prompt-cache read: $0.009 per 1M tokens
- Context window: 131K tokens
- Arena ELO: 1317 as of 2026-10-08
Available on
AWS Bedrock regions: ap-northeast-1, eu-central-1, eu-west-1, us-east-1, us-east-2, us-west-2
Market position
On blended price ($0.0684 per 1M tokens at 30% input, 70% output) it is cheaper than 97% of models in the Budget price tier: 6th cheapest of 164. Across every price tier it is cheaper than 98% of priced text models in the catalogue: 6th cheapest of 308. Its Arena ELO of 1317 ranks 139th of the 156 models that carry one, above 11% of them. With an Arena score of 1317, gpt-oss-20b sits below OptimNow's lowest working quality threshold, Standard (Arena 1415+), so the quality-bar method would not shortlist it for any of the 8 use cases the table prices. Each threshold is a distance below the best ranked model, which scores 1525 on Arena scores as of 8 Oct 2026.
Licence, size and capabilities
gpt-oss-20b is open source: its weights are published under Apache 2.0, an OSI-approved licence with no field-of-use limit, so it can be self-hosted as well as bought by the token. Its published size is 20B parameters. Listed capabilities: Text, Reasoning, Agents. Filling its 131K context window once costs $0.00236 in input tokens at list price.
Pricing by host
A host is a company that runs the model and sells it by the token: the model maker, a cloud platform like Amazon Bedrock, Azure or Vertex, or an independent host. One model, several hosts, and often several prices. This model is sold by 10 different hosts. The price we quote elsewhere on this page is marked. We quote one host's rate so that models stay comparable; it is not a recommendation to buy there.
The price we quote is already the cheapest of the 10 hosts: Darkbloom at $0.018 input / $0.09 output per 1M tokens. It serves a quantized fp8 build, which is not the same product as the full-precision model. The dearest, Groq, charges 3.4x the cheapest.
| Host | Input $/1M | Output $/1M | Precision |
|---|---|---|---|
| Darkbloom (price we quote) | $0.018 | $0.09 | fp8 |
| AkashML | $0.02 | $0.10 | fp4 |
| CoreWeave | $0.03 | $0.13 | fp4 |
| DekaLLM | $0.029 | $0.14 | bf16 |
| DeepInfra | $0.03 | $0.14 | bf16 |
| Parasail | $0.03 | $0.15 | fp4 |
| Amazon Bedrock | $0.07 | $0.15 | not declared |
| SiliconFlow | $0.04 | $0.18 | fp8 |
| $0.07 | $0.25 | not declared | |
| Groq | $0.075 | $0.30 | not declared |
List prices as published on 2026-10-22, via OpenRouter. Negotiated rates, committed-use discounts and provisioned capacity are not shown. These hosts do not all serve the model at the same numeric precision, so the cheapest is not a like-for-like substitute for the dearest.
Monthly cost at 100K requests, by business use case
From the after-discounts figure (prompt caching on the cached input share, batch API for async workloads, only where OpenAI publishes those rates), which assumes you implement both, to the list price. One figure means no discount is published.
| Use case | Tokens (in/out) | Monthly (after discounts – list) |
|---|---|---|
| Support Ticket | 1,500 / 500 | $6 – $7 |
| Knowledge Q&A | 2,000 / 800 | $10 – $11 |
| Meeting Summary | 10,000 / 1,200 | $28 – $29 |
| Marketing Content | 2,500 / 1,800 | $20 – $21 |
| Coding Task | 3,000 / 2,000 | $22 – $23 |
| Invoice Processing | 1,500 / 600 | $8 |
| Call Summary | 2,000 / 700 | $10 |
| Agent Workflow | 6,000 / 3,000 | $34 – $38 |
Price history
gpt-oss-20b entered our daily price archive on 2026-10-18; tracking began on 2026-08-15. It listed at $0.018 input / $0.09 output per 1M tokens that day. No list price change since 2026-10-18. Last checked against the 2026-10-22 snapshot.
Alternatives to gpt-oss-20b
- Next cheaper model in the Budget price tier: Ling 3.0 Flash (inclusionai) at $0.0504 blended, 26% less.
- Next more expensive: Llama 3.1 8B Instruct (Meta) at $0.071 blended, 4% more.
The nearest models in the Budget price tier on Arena ELO, each with its blended price and score difference against gpt-oss-20b.
- Llama 3.3 70B Instruct (Meta): $0.10 in / $0.32 out per 1M, Arena ELO 1318 (271% more expensive, 1 ELO point higher)
- GPT-4o-mini (OpenAI): $0.15 in / $0.60 out per 1M, Arena ELO 1318 (580% more expensive, 1 ELO point higher)
- Nemotron 3 Nano 30B A3B (Nvidia): $0.06 in / $0.24 out per 1M, Arena ELO 1314 (172% more expensive, 3 ELO points lower)
- GPT-4.1 Nano (OpenAI): $0.10 in / $0.40 out per 1M, Arena ELO 1322 (353% more expensive, 5 ELO points higher)
- Cheapest within 15 Arena points: Gemma 3 4B (Google), 31% cheaper, 14 ELO points lower, out of 13 models in that band.
- Highest Arena ELO within 20% of this price: gpt-oss-20b itself has the highest score of the 3 models priced within 20% of it.
Other models from OpenAI
The 8 nearest to gpt-oss-20b in price, of 53 other OpenAI models in the catalogue. Prices per 1M tokens, input then output.
- GPT-5 Nano: $0.05 / $0.40
- GPT-4.1 Nano: $0.10 / $0.40
- GPT-6 Luna: $0.10 / $0.50
- GPT-6 Luna Pro: $0.10 / $0.50
- GPT-4o-mini: $0.15 / $0.60
- gpt-oss-120b: $0.15 / $0.60
- GPT-5.6 Luna: $0.20 / $1.20
- GPT-5.6 Luna Pro: $0.20 / $1.20
All 54 OpenAI models and prices
Common questions about gpt-oss-20b pricing
- How much does gpt-oss-20b cost?
- gpt-oss-20b costs $0.018 per 1M input tokens and $0.09 per 1M output tokens at Darkbloom, the host whose rate we quote. Darkbloom serves a quantized fp8 build, which is not the same product as the full-precision model. Prices as listed on 2026-10-22.
- Which host is cheapest for gpt-oss-20b?
- The price we quote is already the cheapest of the 10 hosts: Darkbloom at $0.018 input / $0.09 output per 1M tokens. It serves a quantized fp8 build, which is not the same product as the full-precision model. The dearest, Groq, charges 3.4x the cheapest. Host rates as listed on 2026-10-22.
- Does gpt-oss-20b offer batch or cached pricing?
- Yes, for prompt caching only. Cache reads cost $0.009 per 1M tokens, as listed on 2026-10-22. No batch pricing is published.
- What would gpt-oss-20b cost per month?
- With the Support Ticket profile (1,500 input and 500 output tokens per request), 100,000 requests a month cost about $7 on gpt-oss-20b at the list price quoted on 2026-10-22.
- Can gpt-oss-20b be self-hosted?
- Yes. gpt-oss-20b is open source: its weights are published under Apache 2.0, an OSI-approved licence with no field-of-use limit, so it can be self-hosted as well as bought by the token.
- Has the price of gpt-oss-20b changed?
- gpt-oss-20b entered our daily price archive on 2026-10-18; tracking began on 2026-08-15. It listed at $0.018 input / $0.09 output per 1M tokens that day. No list price change since 2026-10-18. Last checked against the 2026-10-22 snapshot.
These are list prices. Your real cost also depends on caching and batching, negotiated discounts, marketplace agreements, and the contracts you already have. OptimNow audits exactly that.
Build the business case · Get a FinOps review
LLM pricing · Models by maker · Cloud compute pricing · Compare models · Price barometer · FinOps guides · API and MCP server