DeepSeek V4.1 Flash vs GLM 5.3 Flash
Compare with other modelsThe biggest difference between DeepSeek V4.1 Flash and GLM 5.3 Flash is the unsupported claims. The share of conversations with no unsupported claims is twice as high with DeepSeek: 48% against 24%.
Both models score well on tool use, and GLM does better on task completion. But for a CX agent, DeepSeek's reduction in unsupported claims is worth the small extra cost: about $0.013 per conversation against $0.011.
| Metric | DeepSeek Sep 15, 2026 | Zhipu Sep 10, 2026 |
|---|---|---|
| Overview | ||
| Quality score | 85 Best among selected models #2 of 17 | 81 #7 of 17 |
| Median reply | 5.1s Best among selected models Served by Phala | 6.4s Served by Together |
| Cost per conversation | $0.013 | $0.011 Best among selected models |
| Quality scores | ||
| Tool use | 97 Best among selected models | 96 |
| Task completion | 89 | 95 Best among selected models |
| Context retention | 81 Best among selected models | 75 |
| Grounding | 93 Best among selected models | 90 |
| Safety | 96 Best among selected models | 94 |
| Hallucinations % of conversations with no unsupported claims detected | 48% Best among selected models | 24% |
| Running stats | ||
| p95 reply | 19.6s Best among selected models | 34.9s |
| Consistency % of repeat runs ended the same way | 85% Best among selected models | 82% |
| Output per turn | 450 tokens, median | 600 tokens, median |
Reply times cover the model call only. Cost estimates use token usage at list prices.
This whole report is one Voxli workspace: simulated customers, assertion checks, and a frozen, versioned test set that reruns when new models ship.
Test set v3f-2026-09 · edition 2026-09-08