← Scoreboard
GLM 5.3
Each provider gets the same short prompts. We count how often its top answer is the same as Z.ai's own.
Tested
10
of 47 on OpenRouter
Match
8
Watching
2
0 no match
Typical agreement
72%
same answer as the maker
| Provider | Same answer as the maker · typical | Result | Precision | Price out / 1M |
|---|---|---|---|---|
| Z.AI · fp8 | makersame answer twice 62% of the time | Reference | fp8 | $4.40 |
| Novita · fp8 | 75%76 of 102 prompts · typical 71% | Matches | fp8 | $1.98 |
| Makora · fp4 | 74%56 of 76 prompts · typical 71% | Matches | fp4 | $2.00 |
| Decart · fp4 | 74%75 of 102 prompts · typical 71% | Matches | fp4 | $1.78 |
| GMICloud · fp8 | 72%73 of 101 prompts · typical 71% | Matches | fp8 | $2.20 |
| Reka | 71%72 of 101 prompts · typical 72% | Matches | unknown | $2.10 |
| SiliconFlow · fp8 | 70%71 of 102 prompts · typical 72% | Matches | fp8 | $2.20 |
| Wafer · us | 58%59 of 101 prompts · typical 72% | On watch | fp4 | $1.40 |
| Wafer · fp4 | 56%57 of 102 prompts · typical 72% | On watch | fp4 | $1.40 |
| Tenstorrent · fp8offline | 45%9 of 20 prompts · typical 60% | Matches | fp8 | $2.50 |
Question by question
106 questions. Each square is one question. Lit means the same answer as Z.ai. Hover or tap a column.
Results from Nov 11. Short prompts at temperature 0, so settings cannot change the answer. Honest providers still differ a little, so each one is compared with what is typical for the others. Quality checks are not running yet. A provider is marked as no match only after 2 failed checks on different days.