Solar System
broken high fixed version →
Overall0
Score0/8
Tech69
UX / UI0 / 0
Time4 min 22 s
Cost$0.082
Compare →
DeepSeek V4 Pro
DeepSeekReleased 13/08/2026open weights
Overall49/100#18 of 27
Tech80/100#24 of 27
UX / UI22 / 44#23 of 27
Score21.5/40 (24)
Total cost$0.37
Average time4 min 10 s#15 of 27
Total tokens190,929
API price: $0.66 / $1.98 per 1M input / output tokens · as of 16/10/2026 source ↗
What is DeepSeek V4 Pro good for?
DeepSeek's large open-weight Mixture-of-Experts model. According to the maker, the GA version greatly improved agent capabilities.
Description based on the maker · source ↗
- Maker
- DeepSeek
- Released
- 13/08/2026
- Context window
- 1M tokens
- Model weights
- open (MIT licence)
- Reasoning levels
- low high max
How did DeepSeek V4 Pro do in the AI Lab tests?
- 4 of 5 outputs load without errors; 1 more ran after one fix by the same model.
- Its best technical result was on the “GTA-style game” prompt (90/100).
- Did not work (original output): Solar System.
- It implemented the chess rules flawlessly – passing the objective perft test on all 6 positions.
- Average time per prompt 4 min 10 s#15 of 27
- Average cost per prompt $0.075#9 of 27
- Overall score 49/100#18 of 27
- Technical health 80/100#24 of 27
- Automatic UX rating 22/100#23 of 27
Outputs
All outputs of DeepSeek V4 Pro on the 5 prompts every other model got as well.
GTA-style game
loads high
Overall51
Score4/8
Tech90
UX / UI33 / 50
Time2 min 26 s
Cost$0.044
Compare →
Monitor test
loads high
Overall79
Score7.5/8
Tech75
UX / UI0 / 100
Time3 min 05 s
Cost$0.055
Compare →
New York & Central Park map
loads high
Overall44
Score3/8
Tech82
UX / UI76 / 10
Time3 min 24 s
Cost$0.062
Compare →
Chess with an engine
loads high
Overall70
Score7/8
Tech84
UX / UI0 / 60
Time7 min 31 s
Cost$0.13
Compare →
Cost = tokens priced with the API price list valid on the date shown next to the cost (prices change). ≈ = estimate (only the total token count is known), no ≈ = exact from input / output tokens. “Loads” = the page opens without errors; technical health also checks behaviour during use – judge the quality in the viewer.