deepseek-reasoner
DeepSeeklegacy model Its test configurations are retired, so it is no longer run. The figures are from its best legacy runs, scored as they ran at the time; it is not ranked.
- average±11.4 (95%)
- 46.3
- benchmarksof 12
- 1
- per 1,000 documentsmean over its benchmarks
- $2.15
- seconds per inputmean response time
- 147.7
By benchmark
| Benchmark | ScoreThe best legacy run on each benchmark, on the benchmark's metric scaled to 0–100. ± is the 95% sampling margin, from resampling the run's inputs. | RankRank among all current models on that benchmark; ≈ for models that cannot be told apart within sampling error. Legacy models are not ranked. | Per 1,000 docs | s / input | Best run |
|---|---|---|---|---|---|
| Book Advert XML Repair | 46.3±11.4 | – | $2.15$0.11 run | 147.72 | T0764 · 25 Mar 2026prompt.txt |
Not run on: Bibliographic Data, Blacklist Cards, Business Letters, Company Lists, Duty Rosters, Fraktur Adverts, General Meeting Minutes, Library Cards, Magazine Pages, Medieval Manuscripts, Personnel Cards.
Price
| From | Input, $ per million tokens | Output, $ per million tokens | Source |
|---|---|---|---|
| 22 Jul 2026 | 0.14 | 0.28 | archived price page |
| 23 Mar 2026 | 0.28 | 0.42 | archived price page |