Darkbloom model prices: what each model pays per million tokens
Updated
Short answer: Darkbloom publishes every model’s price at api.darkbloom.dev/v1/pricing. As of September 30, 2026, 01:41 UTC, output prices ran from $0.09 per million tokens (gpt-oss-20b) to $2.20 (Qwen3.8 27B), and the provider keeps all of it, because Darkbloom’s platform fee is 0%. Prices change, so check the live endpoint before you decide anything.
Prices as of September 30, 2026, 01:41 UTC
US dollars per million tokens, from api.darkbloom.dev/v1/pricing, for the models listed in Darkbloom’s capacity endpoint at the same time. Sorted by output price.
| Model ID | Input | Output | Cached input |
|---|---|---|---|
| EigenLabs/Qwen3.8-27B-4bit-mtp | $0.05 | $2.20 | $0.025 |
| qwen3.5-35b-a3b | $0.08 | $0.75 | $0.04 |
| qwen3.6-35b-a3b-vl-mtp-mxfp8 | $0.05 | $0.70 | $0.025 |
| ternary-bonsai-2-27b | $0.075 | $0.50 | $0.0375 |
| gemma-4-26b-qat-4bit | $0.042 | $0.22 | $0.021 |
| gemma-4-26b-8bit | $0.042 | $0.22 | $0.021 |
| nvidia-nemotron-3.5-lightning | $0.039 | $0.18 | $0.0195 |
| Qwen3.5-9B | $0.08 | $0.13 | $0.04 |
| gpt-oss-20b | $0.018 | $0.09 | $0.009 |
Also priced, but not being served then
- gemma-4-26b: $0.042 input, $0.22 output, $0.021 cached.
- mimo-v2.6-flash-mopd: $0.07 input, $0.28 output, $0.002 cached.
- qwen3.8-flash-next: $0.13 input, $0.40 output, $0.065 cached.
- qwen3-vl-30b-a3b-instruct: $0.09 input, $0.40 output, $0.045 cached.
- Fallback for a model without its own row: $0.05 input, $0.20 output, $0.025 cached.
What the columns mean
- Input: each prompt token the model reads, per million.
- Output: each token the model writes, per million. Depending on the model it is priced from about 1.6 times (Qwen3.5-9B) to 44 times (Qwen3.8 27B) the input price.
- Cached input: prompt tokens your Mac already held in its prefix cache from an earlier request. They are billed at this lower rate, which is half the input price for most models.
- In the raw JSON, input_price, output_price and cache_read_price are in millionths of a dollar per million tokens: 50000 means $0.05. The input_usd, output_usd and cache_read_usd fields show the same in dollars.
What a request pays you
A request pays (prompt tokens − cached tokens) × input price, plus cached tokens × cached price, plus output tokens × output price, all divided by one million. The provider keeps all of it, because Darkbloom’s global platform fee is currently 0%. A request from your own account to your own Mac pays $0. Example (a made-up request): 2,000 prompt tokens and 500 output tokens on gpt-oss-20b pay (2,000 × $0.018 + 500 × $0.09) ÷ 1,000,000 = $0.000081. The same request on Qwen3.8 27B pays $0.0012.
A higher price doesn’t mean more pay
What your Mac earns depends on how many requests arrive for its model and how many other Macs serve it, not only on the price (see which model to run and how Darkbloom routes requests). Darkbloom also sells through OpenRouter, where the default routing favors the cheapest providers, so Darkbloom’s price against other providers affects how much work reaches the network at all.
Prices change
Darkbloom sets these prices and has changed them over time. On September 29, 2026, the table on Darkbloom’s about page still showed older numbers than the endpoint. Check the live list before you rely on any of the numbers above: curl https://api.darkbloom.dev/v1/pricing
Sources
- Live prices: api.darkbloom.dev/v1/pricing
- Models being served: api.darkbloom.dev/v1/models/capacity
- Price units, formulas and constants: github.com/Layr-Labs/d-inference/blob/master/docs/reference/pricing-model.md
- Billing (cached tokens, provider payout): github.com/Layr-Labs/d-inference/blob/master/docs/architecture/billing.md
How BloomGauge helps
BloomGauge’s network view shows public Darkbloom traffic, capacity and pricing per model, and on your own Mac it shows confirmed credits per model and hour, so you can see what a price actually turns into.
Questions
How much does Darkbloom pay per token?
It depends on the model. As of September 30, 2026, output prices ranged from $0.09 per million tokens (gpt-oss-20b) to $2.20 (Qwen3.8 27B), and input prices from $0.018 to $0.08 for the models being served. Providers keep the full price because the platform fee is 0%.
Where can I see current Darkbloom prices?
At https://api.darkbloom.dev/v1/pricing. It lists every model’s input, output and cached-input price, in dollars per million tokens.
Do Darkbloom providers get paid for cached tokens?
Yes, at the cached-input price, which is half the input price for most models. The prompt tokens that weren’t cached are paid at the full input price.
Related
- How Darkbloom pays Mac providers: token earnings and base rewards
- Which Darkbloom model should I run on my Mac?
- Where Darkbloom’s paid requests come from: OpenRouter and the Darkbloom API
- How much can a Mac earn on Darkbloom?
Updated 2026-09-29. Still stuck? Ask in #bloomgauge on the Darkbloom Slack or contact us. BloomGauge is independent and not affiliated with Darkbloom.