Search Intelligence Score
54.3
+35.9 lift from search
050100
- Lift from search
- +35.9
- 18.4 without search
- Cost per 1K tasks
- $66.6
- Below the median score of 64.4, so not ranked on cost
Accuracy by benchmark
Accuracy on each eval, with search (solid) and without (light).
- DSQA69.9 vs 26.1
- HLE40.0 vs 23.0
- WISER53.0 vs 6.0
Cost per 1K tasks
How the cost splits between model inference, search calls, and page extraction.
- Inference
- $54.9
- 82%
- Search
- $4.32
- 6%
- Extract
- $7.40
- 11%
Compared with
Other models from the same provider alongside models that score closest.
- GLM-5.254.3Cost $66.6 · Lift +35.9
- GLM 5.3 Flash60.8Cost $16.6 · Lift +40.4
- GLM 5.362.7Cost $78.5 · Lift +40.8
- Nem 3 Ultra53.0Cost $113 · Lift +37.7
- Beam56.0Cost $161 · Lift +27.2
| Model | Score | Lift | Cost per 1K tasks |
|---|---|---|---|
| GLM-5.2 | 54.3 | +35.9 | $66.6 |
| GLM 5.3 FlashGLM 5.3 Flash | 60.8 | +40.4 | $16.6 |
| GLM 5.3GLM 5.3 | 62.7 | +40.8 | $78.5 |
| Nem 3 UltraNemotron 3 Ultra 550B | 53.0 | +37.7 | $113 |
| BeamBeam | 56.0 | +27.2 | $161 |
Search Intelligence Score averages accuracy on DSQA, HLE, and WISER with equal weight; cost is averaged the same way, with cost shown per 1,000 tasks. Evaluated . Leaderboard last updated . Full methodology.
# GLM-5.2 · Search Capability Leaderboard
- Search Intelligence Score: 54.3 (#43 in Search Intelligence)
- Without search: 18.4 · Lift from search +35.9
- Cost per 1K tasks: $66.6 (below the median score of 64.4, so not ranked on cost)
- Evaluated:
- Lab: Z.ai · Weights: Open weights · Model id: z-ai/glm-5.2
## Accuracy by benchmark
| Suite | With search | Without search |
|---|---|---|
| DSQA | 69.9 | 26.1 |
| HLE | 40.0 | 23.0 |
| WISER | 53.0 | 6.0 |
## Compared with
| Model | Score | Lift | Cost per 1K tasks |
|---|---|---|---|
| GLM-5.2 | 54.3 | +35.9 | $66.6 |
| GLM 5.3 Flash | 60.8 | +40.4 | $16.6 |
| GLM 5.3 | 62.7 | +40.8 | $78.5 |
| Nemotron 3 Ultra 550B | 53.0 | +37.7 | $113 |
| Beam | 56.0 | +27.2 | $161 |
Leaderboard last updated . Methodology: /leaderboard#methodology