gpt-oss-120b
$0.15 per 1M input tokens and $0.60 per 1M output tokens · 131K context · tools, reasoning, JSON, open weights.
Prices checked Oct 21, 2026 · tracked by LLM Cost Index since Oct 21, 2026
Cheaper ways to run the same model: gpt-oss-120b (batch) at $0.03 / $0.136 — same model, different service (e.g. batch processing with a delay).
What gpt-oss-120b costs per month
Estimated monthly bill for typical workloads, using the cached-input price for repeated context. Try your own numbers in the cost calculator.
| Workload | Per month |
|---|---|
| Customer chatbot 1,000 conversations a day, 8 turns each | $88.20 |
| Document Q&A (RAG) 500 questions a day over retrieved documents | $17.32 |
| Coding / tool-using agent 200 tasks a day, long tool-heavy contexts | $48.15 |
| Content generation 30 long articles a day | $1.75 |
| Classification / extraction 100,000 short items a day | $162 |
Price history
We started tracking gpt-oss-120b on Oct 21, 2026; the chart appears as soon as there is a second data point. See all recent price changes.
Where to run gpt-oss-120b: 24 providers
AkashML is the cheapest at $0.03 input / $0.187 output per 1M tokens. Uptime is the average over the last 30 days as reported by OpenRouter; see provider reliability.
| Provider | Input / 1M | Output / 1M | Context | Uptime | Notes |
|---|---|---|---|---|---|
| AkashML | $0.03 | $0.187 | 131K | 100% | bf16 |
| CoreWeave | $0.03 | $0.17 | 131K | 98.5% | fp4 |
| DekaLLM | $0.03 | $0.18 | 131K | 100% | bf16 |
| Near AI | $0.03 | $0.15 | 131K | 99.9% | fp4 |
| Venice | $0.03 | $0.15 | 128K | 98.4% | int4 |
| DeepInfra | $0.037 | $0.17 | 131K | 100% | bf16 |
| Crusoe | $0.05 | $0.25 | 131K | 100% | bf16 |
| Mancer 2 | $0.05 | $0.25 | 131K | 99.9% | fp8 |
| Novita | $0.05 | $0.25 | 131K | 99.8% | fp4 |
| DigitalOcean | $0.06 | $0.42 | 128K | 98.7% | — |
| $0.09 | $0.36 | 131K | 99.9% | — | |
| BaseTen | $0.10 | $0.50 | 128K | 99.6% | fp4 |
| BaseTen | $0.10 | $0.50 | 128K | 99.8% | fp4 |
| Parasail | $0.10 | $0.75 | 131K | 99.4% | fp4 |
| SambaNova | $0.14 | $0.95 | 131K | 100% | — |
| Amazon Bedrock | $0.15 | $0.60 | 131K | 100% | — |
| Amazon Bedrock | $0.15 | $0.60 | 131K | 100% | — |
| Groq | $0.15 | $0.60 | 131K | 100% | — |
| Mara | $0.15 | $0.75 | 131K | 89.5% | — |
| Nebius | $0.15 | $0.60 | 131K | 100% | fp4 |
| Phala | $0.15 | $0.60 | 131K | 100% | — |
| SiliconFlow | $0.15 | $0.60 | 131K | 91.6% | fp8 |
| Together | $0.15 | $0.60 | 131K | 97.3% | — |
| Cerebras | $0.35 | $0.75 | 131K | 100% | fp16 |
Cheaper alternatives to gpt-oss-120b
Models with the same tool calling, structured output for at least 20% less (3:1 input/output mix), most capable first.
| Model | Input / 1M | Output / 1M | Context | Capabilities |
|---|---|---|---|---|
| Claude Haiku 5.5 Anthropic | $0.10 | $0.50 | 1M | visiontoolsreasoningJSON |
| GLM 5.3 Flash Z.ai | $0.041 | $0.12 | 1.05M | visiontoolsreasoningJSONopen weights |
| Ling 3.1 Flash inclusionAI | $0.075 | $0.225 | 262K | toolsreasoningJSON |
| GPT-6 Luna OpenAI | $0.10 | $0.50 | 1.05M | visiontoolsreasoningJSON |
| MiMo-V2.6-Flash Xiaomi | $0.14 | $0.28 | 1.05M | visionaudiotoolsreasoningJSONopen weights |
How to pay less for gpt-oss-120b
For a customer chatbot (1,000 conversations a day, 8 turns each) that costs $104 a month at list price without caching:
| Option | Per month | You save |
|---|---|---|
| Switch to gpt-oss-120b (batch) | $22.08 | $82.32 (79%) |
| Route to Near AI | $23.40 | $81.00 (78%) |
| Move to Claude Haiku 5.5 (similar capabilities) | $58.56 | $45.84 (44%) |
| Use prompt caching (60% of input cached) | $88.20 | $16.20 (16%) |
Other workloads: cut-your-bill tool for gpt-oss-120b.
Specifications
| Model ID | openai/gpt-oss-120b |
| Input | text |
| Output | text |
| Tool calling | Yes |
| Structured output | Yes |
| Reasoning mode | Yes |
| Open weights | Yes — openai/gpt-oss-120b |
| Knowledge cutoff | 2024-06-30 |
| Released | Aug 5, 2025 |
FAQ
- How much does gpt-oss-120b cost?
- $0.15 per million input tokens and $0.60 per million output tokens; cached input costs $0.075 per million. Prices as of Oct 21, 2026.
- What is the context window of gpt-oss-120b?
- 131K tokens, with up to 66K output tokens per response.
- Does gpt-oss-120b support images, tools and structured output?
- Images as input: no. Tool calling: yes. Structured (JSON) output: yes. Reasoning mode: yes.
- What are cheaper alternatives to gpt-oss-120b?
- Models with the same key capabilities for at least 20% less include Claude Haiku 5.5 ($0.10 / $0.50), GLM 5.3 Flash ($0.041 / $0.12), Ling 3.1 Flash ($0.075 / $0.225).
More from OpenAI: all OpenAI models · how we collect prices