Open-weight break-even calculator
Enter your tokens per month. See what a closed API, the same open-weight model on a hosted API, and your own rented GPUs cost, and the volume where the GPUs start to win. Every input is visible and every price links its source.
| Option | Per month | Per 1M tokens |
|---|---|---|
| Closed API GPT-6.1 Sol | $280 | $2.80 |
| Hosted open-weight APICheapest gpt-oss-120b on Together AI | $19.50 | $0.20 |
| Rented GPUs 1 GPU at $4.29/hr, always on | $3,932 | $39.32 |
- Rented GPUs cost less than GPT-6.1 Sol from 1.4B tokens a month.
- Rented GPUs never cost less than gpt-oss-120b on Together AI at these settings.
- GPU bill: 1 GPU x $4.29 x 730 hours = $3,132, plus $800 of engineer time. These GPUs can serve about 7.1B tokens a month at 30% load.
Closed API prices from each vendor's pricing page, checked 2026-10-21. Together AI and Lambda prices checked 2026-10-14. US prices before tax. Compare models of similar quality on your own tests before you trust a cost win. It runs in your browser. Nothing you type is sent anywhere.
Where the numbers come from
- Closed API prices come from the official OpenAI, Anthropic, and Google pricing pages, the same file our other AI calculators use.
- Hosted open-weight prices come from the serverless table on Together AI's pricing page, checked 2026-10-14.
- GPU prices are on-demand rates per GPU per hour from Lambda's pricing page, checked 2026-10-14.
- Throughput is measured by SemiAnalysis InferenceX on real GPUs, at 50 tokens per second per user. It counts input plus output tokens.
What the calculator assumes
- GPUs stay on all month, 730 hours. You pay for idle hours.
- Real traffic peaks in work hours, so 30% average load is our default. Set your own.
- Engineer time defaults to 8 hours a month at $100, which is what we plan for. Set your own.
- No caching, batch, or off-peak discounts on any API. Storage, egress, and setup time are left out.
- A cheaper option only counts if the model passes your own tests. The models are not equal in quality.
Default scenario at four volumes
| Tokens a month | Closed API | Hosted open-weight | Rented GPUs |
|---|---|---|---|
| 10M | $28.00 | $1.95 | $3,932 1 GPU |
| 100M | $280 | $19.50 | $3,932 1 GPU |
| 1B | $2,800 | $195 | $3,932 1 GPU |
| 10B | $28,000 | $1,950 | $7,063 2 GPUs |
Break-even volume for each GPU setup
| GPU setup | GPU rent a month | vs GPT-6 Astra | vs GPT-6.1 Sol | vs GPT-6 Luna | vs same model on Together |
|---|---|---|---|---|---|
| gpt-oss-120b on 1x H100 $4.29/GPU/hr, 8,973 tokens/s per GPU | $3,132 | 281M | 1.4B | never | never 20B at 100% load |
| gpt-oss-120b on 1x B200 $6.99/GPU/hr, 44,489 tokens/s per GPU | $5,103 | 422M | 2.1B | never | 30B 30B at 100% load |
| Kimi K3 on 16x B200 $9.86/GPU/hr, 6,505 tokens/s per GPU | $115,165 | 8.3B | 41B | never | 28B 28B at 100% load |
Every price the calculator uses
| Item | Price or rate | Source |
|---|---|---|
| gpt-oss-120b, hosted | $0.15 in / $0.60 out per 1M tokens | Together AI |
| Gemma 4 31B, hosted | $0.39 in / $0.97 out per 1M tokens | Together AI |
| GLM-5.3-Flash, hosted | $0.15 in / $0.50 out per 1M tokens | Together AI |
| DeepSeek V4.1 Flash, hosted | $0.30 in / $1.20 out per 1M tokens | Together AI |
| DeepSeek V4 Pro 0813, hosted | $1.32 in / $3.96 out per 1M tokens | Together AI |
| GLM-5.3, hosted | $1.40 in / $4.40 out per 1M tokens | Together AI |
| Qwen3.8-2.4T-A95B (Max), hosted | $2.00 in / $6.00 out per 1M tokens | Together AI |
| Kimi K3, hosted | $3.00 in / $15.00 out per 1M tokens | Together AI |
| gpt-oss-120b on 1x H100 NVIDIA H100 SXM 80 GB, Lambda 1x instance | $4.29 per GPU-hour, 8,973 tokens/s per GPU | Lambda , InferenceX |
| gpt-oss-120b on 1x B200 NVIDIA B200 SXM6 180 GB, Lambda 1x instance | $6.99 per GPU-hour, 44,489 tokens/s per GPU | Lambda , InferenceX |
| Kimi K3 on 16x B200 NVIDIA HGX B200, Lambda 1-Click Cluster of 16 GPUs | $9.86 per GPU-hour, 6,505 tokens/s per GPU | Lambda , InferenceX |
Need help choosing and testing a model for your product? See our AI integration work.
Questions clients ask
Short answers, no sales language. If yours isn't here, ask an engineer directly.
Ask us anythingWhen is self-hosting an open-weight model cheaper than an API?
Only at high, steady volume. With gpt-oss-120b on one rented H100 at $4.29 an hour, 30% average load, and 8 engineer hours a month, the GPU costs less than GPT-6.1 Sol from about 1.4 billion tokens a month. Against the same gpt-oss-120b on a hosted API at $0.15 / $0.60, the H100 never catches up at 30% load. Prices checked 2026-10-14.
How much does it cost to rent a GPU for an LLM?
Lambda lists one H100 SXM at $4.29 an hour and one B200 at $6.99 an hour, on demand, checked 2026-10-14. Left on all month (730 hours), that is about $3,132 and $5,103.
Why does GPU load matter so much?
A rented GPU bills every hour, busy or idle. Throughput figures are measured at full load. If your GPU is busy 30% of the time, each token costs over three times what it would at full load.
Does the calculator store what I type?
No. It runs in your browser. Nothing you enter is sent anywhere.
More free tools
- AI crawler checker Check whether ChatGPT, Perplexity, Claude, Google, and Bing can read a page: robots.txt, firewall blocks, JavaScript, and noindex.
- AI model compare Put 2 to 4 GPT, Claude, and Gemini models side by side: price, context window, retirement date, and your monthly cost.
- AI model picker Pick the job, volume, and prompt length. Get three API models to test, each with a monthly cost and a test to run first.
- AI feature cost calculator Pick summarise, extract, chat, or agent, enter users and a model, and see the GPT, Claude, or Gemini cost per user per month.
Not sure which model or setup fits your product?
We add AI features to existing software and test models on your own data before you commit. An engineer replies, not a sales team.