Hunterson Standard Inference, Europe, 3 November 2026
Tuesday 3 November 2026, published 14:37 UTC
Input
$0.2246
Output
$0.9283
per million tokens, input and output
The first assessment.
Hunterson Standard Inference, Europe, 3 November 2026: input $0.2246, output $0.9283 per million tokens. https://hunterson.org/standard/2026-11-03/
- Blended
- $0.266016 input tokens to 1 output token
- Contributors
- 11 of 16providers tested
- Model
- DeepSeek V4.1-Flashthe reference, as the lab published it
- Bought
- 14:05 to 14:36 UTCthrough OpenRouter, from London
- Before discounts
- $0.2786 input, $1.1100 outputdiscounts take off 16.37% of the output price
The price of AI inference, measured by buying it. Each weekday between 14:00 and 15:00 UTC, from London, Hunterson buys the same test job from every provider serving DeepSeek V4.1-Flash through one marketplace, and keeps the receipts. The figures are the averages of what the middle half of qualifying providers charged, per million input and per million output tokens. Standard covers models that use 30 billion active parameters or fewer for each token. A token is about three quarters of a word.
How the price is assessedThe providers behind this figureThe evidence
The leading challenger
DeepSeek V4.1-Flash-1017 has the most providers after the reference. Assessed in the same way today: input $0.2020, output $0.8169 per million tokens, from 9 of 15 contributors; its output price is 0.88 times the reference's. Shown for information: it is not an assessment of the grade, and it does not change the figures above.
History
The chart needs JavaScript. Every figure is in the table of recent assessments and in the full data.
The providers behind this figure
16 providers were tested. 11 passed every test and contributed; the price is the average of the middle 7, leaving out the lowest 2 and the highest 2. 5 were excluded, for the reason shown.
| Provider | Input | Output | Result |
|---|---|---|---|
| BaiduLeft out: lowest quarter | $0.0330 | $0.1320 | Left out of the averageLowest quarter of prices |
| DeepInfraLeft out: lowest quarter | $0.1400 | $0.4200 | Left out of the averageLowest quarter of prices |
| AtlasCloudIn the average | $0.1350 | $0.5400 | In the average |
| StreamLakeIn the average | $0.1470 | $0.5880 | In the average |
| InferenceNetIn the average | $0.1030 | $0.6000 | In the average |
| MakoraIn the average | $0.2600 | $1.1700 | In the average |
| CrusoeIn the average | $0.2900 | $1.2000 | In the average |
| NebiusIn the average | $0.3000 | $1.2000 | In the average |
| ParasailIn the average | $0.3000 | $1.2000 | In the average |
| SiliconFlowLeft out: highest quarter | $0.3000 | $1.2000 | Left out of the averageHighest quarter of prices |
| VeniceLeft out: highest quarter | $0.3000 | $1.2000 | Left out of the averageHighest quarter of prices |
| CoreWeaveExcluded: Complete: 0 of 3 requests completed (HTTP 429) | ExcludedComplete: 0 of 3 requests completed (HTTP 429) | ||
| GMICloudExcluded: Fast enough: median first token 3.76 seconds, limit 3.00 | ExcludedFast enough: median first token 3.76 seconds, limit 3.00 | ||
| Mancer 2Excluded: Reliable: completed requests in 6 of the previous 10 windows, at least 8 of the last 10 needed | ExcludedReliable: completed requests in 6 of the previous 10 windows, at least 8 of the last 10 needed | ||
| MorphExcluded: Fast enough: median first token 3.11 seconds, limit 3.00 | ExcludedFast enough: median first token 3.11 seconds, limit 3.00 | ||
| NovitaExcluded: Correct: no answer, only reasoning text | ExcludedCorrect: no answer, only reasoning text |
Prices are what each provider's receipts show was paid, in US dollars per million tokens.
Recent assessments
| Date | Input | Output | Blended | Contributors |
|---|---|---|---|---|
| 3 November 2026 | $0.2246 | $0.9283 | $0.2660 | 11 of 16 |