As of August 20, 2026, GPT OSS 20B is 95% cheaper per million tokens than o3-mini. GPT OSS 20B is $0.05 / $0.20 per million input/output tokens versus o3-mini at $1.10 / $4.40. Source: https://developers.openai.com/api/docs/pricing
Source: OpenAI API pricing (https://developers.openai.com/api/docs/pricing)
GPT OSS 20B
OpenAI
Input
$0.05
Output
$0.20
o3-mini
OpenAI
Input
$1.10
Output
$4.40
Token pricing, specs, and benchmarks side by side
GPT OSS 20Bo3-miniVerdict
GPT OSS 20B offers significantly lower pricing, while o3-mini leads on benchmark performance. Your choice depends on whether cost efficiency or raw capability matters more for your use case.
As of August 20, 2026, GPT OSS 20B is 95% cheaper on blended LLM API pricing. GPT OSS 20B costs $0.05 per million input tokens and $0.20 per million output tokens ($0.25 blended 1M-in + 1M-out). o3-mini costs $1.10 / $4.40 per million tokens ($5.50 blended). Sources: https://developers.openai.com/api/docs/pricing and https://developers.openai.com/api/docs/pricing. The cheapest LLM for your app still depends on how many output tokens you generate.
GPT OSS 20B OpenAI API pricing is $0.05 input and $0.20 output per million tokens via OpenAI. o3-mini OpenAI API pricing is $1.10 input and $4.40 output per million tokens via OpenAI. Use those four numbers, not ChatGPT Plus or Claude Pro subscription prices, when you are comparing APIs.
o3-mini has a published SWE-bench Verified score of 49.3%. GPT OSS 20B does not have that eval on this page yet.
GPT OSS 20B supports a 131K token context window, while o3-mini supports 200K tokens. o3-mini offers a larger context window. Filling a larger window bills more input tokens, so the cheaper-per-million model can still cost more on long documents.
o3-mini leads on 9 of 12 shared benchmarks versus GPT OSS 20B's 3 wins. Check the comparison table for GPQA Diamond, SWE-bench, MMLU, HLE, and the other evals we track. Token price and benchmark score together are the usual LLM comparison, not either number alone.
Both GPT OSS 20B and o3-mini are production API models. For cost-sensitive production traffic, GPT OSS 20B has better economics on the blended 1M-in + 1M-out scale. For maximum capability, weight the benchmark table for your domain. Many production stacks route cheap models for drafts and a frontier model for the hard turn, which is usually cheaper than sending everything to the expensive API.
Yes. Call GPT OSS 20B through OpenAI and o3-mini through OpenAI with separate API keys, or through a gateway that already wraps both. Keep prompts in tokens, not characters, when you estimate the invoice. AnotherWrapper templates can swap providers without rewriting auth and billing.
List rates are the latest published OpenAI API pricing and OpenAI API pricing, quoted per million tokens. Official rate sheets: https://developers.openai.com/api/docs/pricing and https://developers.openai.com/api/docs/pricing. Prompt caching, batch APIs, and committed-use discounts can change the invoice. Benchmark scores are from official publications and independent evals, updated as new numbers land.
GPT OSS 20B supports up to 33K output tokens per request, while o3-mini supports up to 100K output tokens. Output tokens are usually the expensive half of API pricing, so a higher max output cap is a capability, not a discount.
GPT OSS 20B runs at approximately 1000 tokens/second while o3-mini runs at 115 tokens/second. GPT OSS 20B is faster in raw throughput. Throughput (tokens per second) and time-to-first-token are why two models with the same per-million-token price can still feel different in chat and agents.
Use the model pickers at the top of this LLM comparison. Search any model in the index and the URL updates. Open GPT OSS 20B or o3-mini from the sidebar for the single-model API pricing page, or go back to the LLM pricing table to sort by cheapest blended cost.
This comparison covers API pricing, context window, throughput, and benchmark scores. GPT OSS 20B costs $0.05/M input and $0.20/M output. o3-mini costs $1.10/M input and $4.40/M output.
Rank both on the LLM leaderboard, SWE-bench, and Humanity's Last Exam.
The index
280 models across 33 providers. Search or jump to a lab — every model page stays linked here.
Next step
Auth, billing, and the API layer are already decided. Fork 8 finished AI apps, or have us build the first version with you.
