One job, one card, one API
Qwen: Qwen3.5-35B-A3B on GeForce RTX 3090
The sums below use the “Chatbot for a small team” example: 4,200 jobs a month, 1,500 tokens in, 400 out. 8 people, 25 questions on each of 21 working days. 1,500 tokens in (the question plus a short history) and 400 tokens out. A worked example, not a log from a real team.
Not enough yet
The API comes to £1.91 a month for this job. There is no speed for running it locally, so the electricity and the break-even are left blank.
Memory. Qwen: Qwen3.5-35B-A3B against GeForce RTX 3090's 24 GB, estimated from the Hub's published weight size across the common quants: BF16 71.9 GB no · Q8_0 38.2 GB no · Q6_K 29.7 GB no · Q5_K_M 25.6 GB no · Q4_K_M 21.8 GB yes · IQ4_XS 19.1 GB yes · Q3_K_M 17.5 GB yes · Q2_K 13.5 GB yes. At Q4_K_M, about 21.8 GB, 2.2 GB of the card's 24 GB is left for the context cache. Rule of thumb, not a measured file size.
- API this month
- £1.91
- Electricity this month
- Blank
- GPU purchase
- Not on file
- Break-even
- Needs a speed
- API cost is input tokens and output tokens, priced per million, times jobs per month. No caching, batch discount, or free tier.
- Pound figures use the stored exchange rate. They are not a quote from the company that trained the model.
- No tokens-per-second figure, so the electricity line is blank. Type one in if you have measured it.
- API price: £0.104 in and £0.745 out per million tokens, from $0.140 and $1.00 at 0.74501 dollars to the pound on 2026-11-10. OpenRouter feed 2026-11-11.
- GeForce RTX 3090: 24 GB GDDR6X. Graphics card power: 350 W. Specs read 2026-11-11.
- No sourced speed for this pair, and none was typed in.
- Upper estimate, based on the official graphics card power of 350 W.
OpenRouter prices · Exchange rate · GPU spec · Ofgem unit rate