Solar System
loads medium
Overall64
Score5/8
Tech81
UX / UI85 / 40
Time2 min 18 s
Cost$0.11
Compare →
Gemini 3.6 Flash
GoogleReleased 21/07/2026
Overall48/100#19 of 27
Tech85/100#21 of 27
UX / UI70 / 36#17 of 27
Score19/40 (28.5)
Total cost$0.51
Average time2 min 25 s#10 of 27
Total tokens137,040
API price: $0.75 / $3.75 per 1M input / output tokens · as of 17/10/2026 source ↗
What is Gemini 3.6 Flash good for?
The previous-generation Flash model, balancing speed and multimodal capabilities across everyday and agentic tasks.
Description based on the maker · source ↗
- Maker
- Released
- 21/07/2026
- Model weights
- closed, the model is available through the API only
- Reasoning levels
- low medium high
How did Gemini 3.6 Flash do in the AI Lab tests?
- 3 of 5 outputs load without errors; 2 more ran after one fix by the same model.
- Its best technical result was on the “Monitor test” prompt (100/100).
- Did not work (original output): GTA-style game, Chess with an engine.
- In the objective perft chess test it passed 0 of 6 positions.
- Average time per prompt 2 min 25 s#10 of 27
- Average cost per prompt $0.10#11 of 27
- Overall score 48/100#19 of 27
- Technical health 85/100#21 of 27
- Automatic UX rating 70/100#17 of 27
Outputs
All outputs of Gemini 3.6 Flash on the 5 prompts every other model got as well.
GTA-style game
broken medium fixed version →
Overall0
Score0/8
Tech67
UX / UI0 / 0
Time1 min 52 s
Cost$0.078
Compare →
Monitor test
loads medium
Overall98
Score8/8
Tech100
UX / UI97 / 90
Time2 min 52 s
Cost$0.11
Compare →
New York & Central Park map
loads medium
Overall76
Score6/8
Tech98
UX / UI89 / 50
Time2 min 17 s
Cost$0.100
Compare →
Chess with an engine
broken medium fixed version →
Overall0
Score0/8
Tech80
UX / UI80 / 0
Time2 min 47 s
Cost$0.11
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.