loading…
Market prices/Compare/deepseek-v3.2-vs-gpt-5.6-sol
Live inference pricing, quoted by competing providers on the Omnious market.
One request of 10,000 input and 1,000 output tokens costs $0.0034 on Deepseek V3.2 and $0.088 on GPT 5.6 Sol. That is 96% less on Deepseek V3.2.
A worked example at the best asks above. Your own input:output ratio moves the figure.
Deepseek V3.2 is 99% cheaper on output tokens, $0.44 against $33.00 per 1M, best ask to best ask.
GPT 5.6 Sol ranks higher on the independent quality index, #1 against #69.
Deepseek V3.2 starts answering sooner, 1.7s to first token against 2.1s at the median.
Prices are USDC per 1M tokens unless the row says otherwise. A dot marks the cheaper, faster or higher-ranked cell. Figures the market has not produced read n/a.
| Metric | Deepseek V3.2 | GPT 5.6 Sol |
|---|---|---|
| Best ask · inputper 1M tokens, USDC | $0.30 (better) | $5.50 |
| Best ask · outputper 1M tokens, USDC | $0.44 (better) | $33.00 |
| Cost of one request10,000 input + 1,000 output tokens, at the best asks above | $0.0034 (better) | $0.088 |
| Blended · 3:1 input:outputper 1M tokens at three input tokens per output token | $0.33 (better) | $12.38 |
| Cleared p50 · outputmedian price settled requests actually paid | n/a | n/a |
| Time to first token · p50router-measured median | 1.7s (better) | 2.1s |
| Time to first token · p95the tail, not the typical case | 2.2s (better) | 2.3s |
| Output throughput · p50tokens per second | 51.5 | 120.4 (better) |
| Fill rateserved / attempted | 100.0% (better) | 50.0% |
| Live quote depthasks on the book now | 1 | 1 |
| Providers competingdistinct providers that served in the window | n/a | n/a |
| Context · minsmallest window across live quotes | 131K | 131K |
| Context · maxlargest window across live quotes | 131K | 131K |
| Cached inputpublished cache rate, per 1M tokens | n/a | n/a |
| Issuer | DeepSeek | OpenAI |
| Quality percentilereviewed independent general-quality index | 78.8 | 100.0 (better) |
| Quality rankposition in the same reviewed cohort of 322, 1 = best | #69 | #1 (better) |
Every anonymized ask quoted for each class right now. The best ask in the table above is the bottom of these ladders.
Ask side · 1 live quote · 1 price level
Highest price at the top, best ask at the bottom. Bar width is the number of live quotes at or below that price. Provider identities stay private.
Ask side · 1 live quote · 1 price level
Highest price at the top, best ask at the bottom. Bar width is the number of live quotes at or below that price. Provider identities stay private.
Each class across seven task shapes, from routing the router has observed. An axis with no observation is left open and labelled N/A, never drawn as a zero.
Fit 51 · priors only
0 observations
One axis has no observation for Deepseek V3.2, so it is left open instead of plotted at zero.
Fit 85 · priors only
0 observations
Omnious has no price list. Providers quote input and output rates per million tokens in USDC, and every request is an RFQ: the quotes competing for that model class enter a live scoring auction that weighs quoted price against router-measured latency and reliability.
The winner clears under second-score/3 rules. It is paid a price capped by the next genuine rival's score, so quoting low to win and billing high is structurally impossible. Settlement is pay-as-you-go USDC on the actual token counts of the response, and every response carries a signed, verifiable receipt.
Every figure on these pages is therefore a reading of the market at a point in time: the best asks on the book, and the average cleared prices of requests that settled. They move whenever the market does.
One OpenAI-compatible key reaches both Deepseek V3.2 and GPT 5.6 Sol. Switch between them per request, settle in USDC on actual usage, and keep a signed receipt for each response.
Get an API key Try the chatOr read either model on its own: Deepseek V3.2 pricing·GPT 5.6 Sol pricing·all comparisons