Same tokens. Compare cost.
11.38M input + 2.84M output tokens
Frontier · Feb 2026Claude Opus 4.6 | Served by JourierQwen3.8-27B | Best sourced frontierClaude Opus 5.5 | |
|---|---|---|---|
| Token cost | €112.58 | €13.31 | €90.07 |
| Coding scoreSWE-bench Pro | 53.4% | 61.7% | 89.9% |
| Output speedTokens / second | 43Published | ~50Estimated | 92Published |
| Value / € | 0.5× | 4.6× | 1.0× |
Sources. The leader has the highest SWE-bench Pro score among 59 models in LLM Stats. Checked ; refreshed daily. Failed refreshes keep the last valid data.
Opus 4.6: 53.4% · Qwen3.8-27B: 61.7% · Claude Opus 5.5: 89.9%. Opus 4.6 is the February 2026 reference.
Workload. Identical input and output token counts, priced at standard uncached rates. 100% uses a theoretical hour of Jourier’s warm GPU with sufficient concurrency and quota. It is not measured capacity or a per-key guarantee. Completion time differs between models; these prices do not imply that each model processes the selected volume in one hour. Equal token counts do not imply equal task results.
Speed. Output tokens per second during generation, excluding initial waiting time. Published speeds and Jourier’s estimate use different setups; this is not a controlled head-to-head test. The speed observations below are dated separately from the daily frontier refresh. A new leader without a sourced speed shows “Not measured”.
- Claude Opus 4.6: 43.2 output tokens/s. Artificial Analysis measurement of Anthropic’s API, adaptive reasoning at max effort, checked 25 September 2026.
- Jourier · Qwen3.8-27B: 50 output tokens/s. Owner-derived estimate for one warm request at roughly 2k context on Jourier’s RTX PRO 6000. Not a load-test result. A cold start adds 3–5 minutes before generation.
- Claude Opus 5.5: 91.5 output tokens/s. Artificial Analysis measurement of Anthropic’s API, adaptive reasoning at xhigh effort with default fallback, checked 25 September 2026.
Rates per million tokens. Input / output:Opus 4.6: $5 / $25Qwen3.8-27B: €0.42 / €3Claude Opus 5.5: $4 / $20
ECB, 24 September 2026: €1 = $1.1367. Taxes and conversion fees not modelled. No batch or caching discounts assumed; these can reduce competitor costs. Jourier charges standard input rates for repeated prompts. Reasoning counts as output.
Value. SWE-bench Pro score ÷ token cost, with the leader set to 1×. This illustrative index excludes speed and waiting time. Published model results use different test setups and are not tests of Jourier’s deployment. This does not measure cost per completed task.
Minimums. Across workspace keys: €1 per Helsinki clock hour containing a request; €5 per month with an active key, even without usage. The displayed Jourier cost includes one €1 hourly minimum. Use spanning additional clock hours can cost more. Token charges count towards these minimums. The monthly floor is not included above.