OpenAI · language model
gpt-oss-120b: pricing, context and benchmarks
gpt-oss-120b is OpenAI's model released on August 5, 2025. It costs $0.15 per million input tokens and $0.6 per million output tokens, with a 131.1K-token context window. Epoch AI ranks it #101 of 125 rated models.
| Performance (ECI) | 8/40 | |
|---|---|---|
| Price | 23/30 | |
| Context | 4/15 | |
| Features | 15/15 |
What you need to know about gpt-oss-120b
OpenAI, based in United States, put gpt-oss-120b online on August 5, 2025. It accepts text as input and produces text. Its context window reaches 131,072 tokens, larger than 29 % of the models in the registry, and it can write up to 65,536 tokens in one go.
It has been available for about 15 months, long enough for its prices and scores to have settled.
gpt-oss-120b pricing
On OpenRouter, gpt-oss-120b costs $0.15 per million input tokens and $0.6 per million output tokens. Counting three tokens read for every token written, its average price is $0.262 per million, more expensive than 25 % of paid models (2.9 times cheaper than the median of $0.76).
A concrete example: a task that sends 2,000 tokens (about 1,500 words) and receives a 500-token answer costs $0.0006. At 1,000 tasks a day, budget $18 a month. The price calculator redoes this with your own volumes.
Tokens already sent and cached are billed at $0.075 per million. With batch processing (delayed answers), the price drops to $0.03 input and $0.14 output.
Performance and benchmarks
The Epoch Capabilities Index (ECI), which aggregates published results across many benchmarks, gives gpt-oss-120b a score of 139.8 (range 135.6 to 142.1). That places it #101 of 125 rated models.
Scores published by Epoch AI on the benchmarks the site tracks:
| Benchmark | What it measures | Score | Date |
|---|---|---|---|
| Terminal-Bench | real tasks in a terminal | 18.7 % | Aug 5, 2025 |
| Aider Polyglot | coding exercises in several languages | 41.8 % | Aug 5, 2025 |
For at least equal performance, DeepSeek V4 Flash 0731, DeepSeek V4 Flash 0423, and Qwen3.7 Flash cost at least 30% less.
Features, privacy and local use
- It supports tool calling, so it can drive an AI agent, call APIs or MCP servers.
- It can answer in an enforced JSON format (structured outputs).
- It is a reasoning model: it "thinks" before answering, which helps on hard tasks but uses more output tokens.
- Its knowledge stops at June 2024.
Its weights are open and downloadable from Hugging Face (apache-2.0 licence): you can run it on your own machines without sending your data to a third party. With 116,829,156,672 billion parameters, it needs about 65,283,378,589 GB of memory at 4-bit quantisation. Check with the VRAM calculator. It was downloaded 2,893,802 times on Hugging Face last month.
Other OpenAI models
| Model | Released | Avg. price | ECI |
|---|---|---|---|
| GPT-6.1 Sol | Sep 29, 2026 | $4 | 168.4 |
| GPT-6.1 Sol Pro | Sep 29, 2026 | $4 | — |
| GPT-6 Sol | Sep 22, 2026 | $4 | 163.7 |
| GPT-6 Luna | Sep 22, 2026 | $0.2 | 156.8 |
| GPT-6 Luna Pro | Sep 22, 2026 | $0.2 | — |
| GPT-6 Sol Pro | Sep 22, 2026 | $4 | — |
| GPT-6 Astra | Sep 4, 2026 | $20 | 168.8 |
| GPT-6 Astra Pro | Sep 4, 2026 | $20 | — |
Verdict
Computed score: 50/100, from public data (performance, price, context, features).
Strengths
- Low price: cheaper than 75 % of paid models.
- Open weights: self-hostable, your data stays with you.
- Agent-ready: tool calling and structured JSON output.
- Discounted batch pricing.
Limitations
- Modest performance: behind 79 % of rated models.
- Text only: reads neither images nor files.
Avoid if…
- you want to host it on a single consumer computer (≈ 65,283,378,589 GB needed at 4-bit).
Alternatives
- DeepSeek V3.1$0.43
- Qwen3 32B$0.13
- Qwen3.5-9B$0.19
- Qwen3 30B A3B$0.22
- Gemma 4 31B$0.25
- DeepSeek V3.2 Exp$0.31
Comparisons and tools
Learn more
Frequently asked questions
How much does gpt-oss-120b cost?
$0.15 per million input tokens and $0.6 per million output tokens on OpenRouter (checked October 25, 2026). A task with 2,000 tokens read and 500 written costs $0.0006.
How big is gpt-oss-120b's context window?
131,072 tokens, roughly 98,000 words, with a maximum of 65,536 output tokens.
Can gpt-oss-120b run locally?
Yes, its weights are published under the apache-2.0 licence. Expect about 65,283,378,589 GB of memory at 4-bit.
Is gpt-oss-120b suitable for an AI agent?
Technically yes: it supports tool calling and structured outputs. Then check its cost for your volume.
How good is gpt-oss-120b?
Epoch AI ranks it #101 of 125 rated models (ECI 139.8). On Terminal-Bench, it scores 18.7%.
What are the alternatives to gpt-oss-120b?
The closest models on price and performance are DeepSeek V3.1, Qwen3 32B, and Qwen3.5-9B.
Related articles
- Mistral AI: models, pricing and agentsOctober 25, 2026
- LLM benchmarks: how to read model rankingsOctober 24, 2026
- Running a local LLM: hardware, software, and limitationsOctober 24, 2026
- Open-source or closed models for your AI agents?October 5, 2026
Neutrality. No affiliate links. No vendor pays to be reviewed, rated or ranked. Reviews are directed by François Chollet, neutral and factual, and every data point is dated and sourced. How we score