GLM-5.2vsNemotron 3 Ultra
Across 4 shared benchmarks, GLM-5.2 leads overall: GLM-5.2 wins 3, Nemotron 3 Ultra wins 1, with 0 ties and an average score difference of +16.26.
GLM-5.2
智谱AI · 2026-06-13 · Reasoning model
Nemotron 3 Ultra
NVIDIA · 2026-06-04 · Reasoning model
GLM-5.23 wins(75%)(25%)1 winNemotron 3 Ultra
Benchmark scores
Grouped by capability, sorted by largest gap within each. 4 shared benchmarks.
General Knowledge
GLM-5.2 2/2| Benchmark | GLM-5.2 | Nemotron 3 Ultra | Diff |
|---|---|---|---|
| LiveBench | 76.249 / 115Normal (No Tools) | 51.7888 / 115Normal (No Tools) | +24.46 |
| HLE | 54.7015 / 181Thinking (With Tools) | 37.4073 / 181Thinking (With Tools) | +17.30 |
AI Agent - Tool Usage
GLM-5.2 1/1| Benchmark | GLM-5.2 | Nemotron 3 Ultra | Diff |
|---|---|---|---|
| Terminal-Bench 2.1 | 8117 / 44Thinking High (With Tools) | 56.4041 / 44Thinking (With Tools) | +24.60 |
Math and Reasoning
Nemotron 3 Ultra 1/1| Benchmark | GLM-5.2 | Nemotron 3 Ultra | Diff |
|---|---|---|---|
| IMO-AnswerBench | 912 / 23Thinking (No Tools) | 92.301 / 23Thinking (With Tools) | -1.30 |
Specs
| Field | GLM-5.2 | Nemotron 3 Ultra |
|---|---|---|
| Publisher | 智谱AI | NVIDIA |
| Release date | 2026-06-13 | 2026-06-04 |
| Model type | Reasoning model | Reasoning model |
| Architecture | MoE | MoE |
| Parameters | 753.33B | 550B |
| Context length | 1M | 1M |
| Max output | 128K | Not available |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | GLM-5.2 | Nemotron 3 Ultra |
|---|---|---|
| Text input | $1.4 / 1M tokens | Not public |
| Text output | $4.4 / 1M tokens | Not public |
| Cache read | $0.26 / 1M tokens | Not public |
One or both models have incomplete public pricing.
Summary
- GLM-5.2leads in:General Knowledge (2/2), AI Agent - Tool Usage (1/1)
- Nemotron 3 Ultraleads in:Math and Reasoning (1/1)
On average across the 4 shared benchmarks, GLM-5.2 scores 16.26 higher.
Largest single-benchmark gap: Terminal-Bench 2.1 — GLM-5.2 81 vs Nemotron 3 Ultra 56.40 (+24.60).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.