Compare / head-to-head
DeepSeek V4 Flashvs
GLM-5.2
DeepSeek V4 Flash leads 11 of 21 shared benchmarks. Both offer a 1M-token context window.
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
11 – 10
DeepSeek V4 Flash leads
Cheaper per token
—
list price, input + output
Larger context
Tie
both 1M tokens
Providers
2 providers
DeepSeek · Z.ai
DeepSeek V4 Flash
DeepSeek · released 2026-07-31
text
- Context
- 1M
- Max out
- —
- Input /1M
- —
- Output /1M
- —
- Cached /1M
- —
- Scores
- 28 · 7 core
Quality
Benchmark matrix
21 shared · 7 only DeepSeek V4 Flash · 25 only GLM-5.2| Benchmark | DeepSeek V4 Flash | GLM-5.2 | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 21 | ||||
| GPQA Diamond | 91% ↗epoch run | 91.9% ↗epoch run | -0.9 pt | GLM-5.2 |
| LMArena Elo | 1436 ↗ | 1475.8 ↗ | -39.8 | GLM-5.2 |
| Chess Puzzles (Epoch AI run) | 33% ↗epoch run | 21% ↗epoch run | +12 pt | DeepSeek V4 Flash |
| DeepSWE v1.1 (Datacurve) | 53.3% ↗ | 43.8% ↗ | +9.5 pt | DeepSeek V4 Flash |
| FrontierMath Tier 4 v2 (Epoch AI run) | 24.4% ↗epoch run | 29.3% ↗epoch run | -4.9 pt | GLM-5.2 |
| FrontierMath Tiers 1-3 v2 (Epoch AI run) | 57.5% ↗epoch run | 59.2% ↗epoch run | -1.7 pt | GLM-5.2 |
| LiveBench Agentic Coding (LiveBench) | 46.8% ↗ | 51.8% ↗ | -5 pt | GLM-5.2 |
| LiveBench Coding (LiveBench) | 75% ↗ | 79.7% ↗ | -4.7 pt | GLM-5.2 |
| LiveBench Data Analysis (LiveBench) | 79.3% ↗ | 73.7% ↗ | +5.6 pt | DeepSeek V4 Flash |
| LiveBench Instruction Following (LiveBench) | 65.5% ↗ | 62.3% ↗ | +3.2 pt | DeepSeek V4 Flash |
| LiveBench Language (LiveBench) | 79.2% ↗ | 76.2% ↗ | +3 pt | DeepSeek V4 Flash |
| LiveBench Mathematics (LiveBench) | 86.8% ↗ | 89.8% ↗ | -3 pt | GLM-5.2 |
| LiveBench Reasoning (LiveBench) | 86.6% ↗ | 78.6% ↗ | +8 pt | DeepSeek V4 Flash |
| LMArena WebDev (LMArena) | 1581 ↗ | 1605 ↗ | -24 | GLM-5.2 |
| Mystery Game Puzzles (Epoch AI run) | 34% ↗epoch run | 19% ↗epoch run | +15 pt | DeepSeek V4 Flash |
| NL2Repo | 54.2% ↗ | 48.9% ↗ | +5.3 pt | DeepSeek V4 Flash |
| OTIS Mock AIME 2024-2025 (Epoch AI run) | 94.4% ↗epoch run | 86.4% ↗epoch run | +8 pt | DeepSeek V4 Flash |
| SimpleBench (SimpleBench) | 61.1% ↗ | 58.8% ↗ | +2.3 pt | DeepSeek V4 Flash |
| SimpleQA Verified | 33.6% ↗epoch run | 34.2% ↗epoch run | -0.6 pt | GLM-5.2 |
| Toolathlon-Verified (HKUST) | 70.7% ↗ | 59.9% ↗ | +10.8 pt | DeepSeek V4 Flash |
| WeirdML (Håvard Tveit Ihle) | 63% ↗ | 70.1% ↗ | -7.1 pt | GLM-5.2 |
| Only DeepSeek V4 Flash reports · 7 | ||||
| Agents' Last Exam | 25.2% ↗ | not reported | — | — |
| AutomationBench Public | 25.1% ↗ | not reported | — | — |
| Cybergym | 76.7% ↗ | not reported | — | — |
| DeepSWE | 54.4% ↗ | not reported | — | — |
| Terminal Bench 2.1 | 82.7% ↗ | not reported | — | — |
| Terminal-Bench 4.0 (Vals AI) | 18.7% ↗ | not reported | — | — |
| Toolathlon-Verified | 70.3% ↗ | not reported | — | — |
| Only GLM-5.2 reports · 25 | ||||
| AIME 2026 | not reported | 90% ↗matharena ⚠ | — | — |
| Agents' Last Exam (ALE-CLI) | not reported | 23.8% ↗ | — | — |
| APEX-Agents (Mercor) | not reported | 45.2% ↗ | — | — |
| AutomationBench v1.0.6 | not reported | 26.2% ↗ | — | — |
| CyberGym | not reported | 77.2% ↗ | — | — |
| DeepSWE v1.1 | not reported | 46.2% ↗ | — | — |
| EBR-bench (Epoch AI run) | not reported | 9.5% ↗epoch run | — | — |
| ExploitBench | not reported | 24.4% ↗ | — | — |
| ExploitGym 2h | not reported | 29 ↗ | — | — |
| ExploitGym 6h | not reported | 39 ↗ | — | — |
| FrontierSWE | not reported | 67.5% ↗ | — | — |
| GDPval-AA v2 | not reported | 1508 ↗ | — | — |
| HLE (with tools) | not reported | 54.7% ↗ | — | — |
| LMArena Agent (LMArena) | not reported | 0.0348 ↗ | — | — |
| PostTrainBench | not reported | 31.7% ↗ | — | — |
| ProgramBench (Almost Solved) | not reported | 9.5% ↗ | — | — |
| SWE-Bench Pro | not reported | 62.1% ↗ | — | — |
| SWE-bench Verified (Epoch AI run) | not reported | 78.7% ↗epoch run | — | — |
| SWE-Marathon v1.1 | not reported | 19.4% ↗ | — | — |
| SWE-rebench 2026-05-15 to 2026-07-01 (Nebius) | not reported | 62.9% ↗ | — | — |
| tau2-bench Banking Knowledge (Sierra) | not reported | 37.1% ↗ | — | — |
| Terminal-Bench 2.1 | not reported | 81% ↗ | — | — |
| Terminal-Bench 3.0 | not reported | 4.6% ↗ | — | — |
| Toolathlon Verified | not reported | 59.9% ↗ | — | — |
| Vending-Bench 2 (Andon Labs) | not reported | 8313.78 ↗ | — | — |
Scores tagged "epoch" or "matharena" are independent runs, used only where the lab has not published its own; ⚠ marks rows MathArena flags as released after the competition. Higher is better on every row. Δ is DeepSeek V4 Flash minus GLM-5.2 in the benchmark's own unit. "Not reported" means the lab has not published that figure; it is not a zero. ↗ opens the source.
Specs & pricing
Side by side
Official pricing pages and model cards| Spec | DeepSeek V4 Flash | GLM-5.2 | Edge |
|---|---|---|---|
| Context window | 1M tokens | 1M tokens | Tie |
| Max output | — | 131K tokens | — |
| Input price / 1M | — | $1.4 ↗ | — |
| Output price / 1M | — | $4.4 ↗ | — |
| Cached input / 1M | — | $0.26 ↗ | — |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | — | $5.8 | — |
| Modalities | text | text | Tie |
| Released | 2026-07-31 | 2026-06-24 | — |
| Cited benchmark scores | 28 | 53 | — |
Reliability
Provider status
Live from /statusMore matchups
DeepSeek V4 Flash vs …
Models sharing the most benchmarksMore matchups
GLM-5.2 vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run DeepSeek V4 Flash and GLM-5.2 on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.