Compare / head-to-head
DeepSeek V4 Flashvs
GLM-5.3
GLM-5.3 leads 16 of 22 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
6 – 16
GLM-5.3 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
22 shared · 6 only DeepSeek V4 Flash · 21 only GLM-5.3| Benchmark | DeepSeek V4 Flash | GLM-5.3 | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 22 | ||||
| GPQA Diamond | 91% ↗epoch run | 90.9% ↗epoch run | +0.1 pt | DeepSeek V4 Flash |
| LMArena Elo | 1436 ↗ | 1478.5 ↗ | -42.5 | GLM-5.3 |
| Agents' Last Exam | 25.2% ↗ | 28.5% ↗ | -3.3 pt | GLM-5.3 |
| Chess Puzzles (Epoch AI run) | 33% ↗epoch run | 21% ↗epoch run | +12 pt | DeepSeek V4 Flash |
| DeepSWE v1.1 (Datacurve) | 53.3% ↗ | 69% ↗ | -15.7 pt | GLM-5.3 |
| FrontierMath Tier 4 v2 (Epoch AI run) | 24.4% ↗epoch run | 29.3% ↗epoch run | -4.9 pt | GLM-5.3 |
| FrontierMath Tiers 1-3 v2 (Epoch AI run) | 57.5% ↗epoch run | 68.8% ↗epoch run | -11.3 pt | GLM-5.3 |
| LiveBench Agentic Coding (LiveBench) | 46.8% ↗ | 60.9% ↗ | -14.1 pt | GLM-5.3 |
| LiveBench Coding (LiveBench) | 75% ↗ | 79% ↗ | -4 pt | GLM-5.3 |
| LiveBench Data Analysis (LiveBench) | 79.3% ↗ | 70.2% ↗ | +9.1 pt | DeepSeek V4 Flash |
| LiveBench Instruction Following (LiveBench) | 65.5% ↗ | 69.3% ↗ | -3.8 pt | GLM-5.3 |
| LiveBench Language (LiveBench) | 79.2% ↗ | 79.9% ↗ | -0.7 pt | GLM-5.3 |
| LiveBench Mathematics (LiveBench) | 86.8% ↗ | 87.9% ↗ | -1.1 pt | GLM-5.3 |
| LiveBench Reasoning (LiveBench) | 86.6% ↗ | 85.8% ↗ | +0.8 pt | DeepSeek V4 Flash |
| LMArena WebDev (LMArena) | 1581 ↗ | 1623 ↗ | -42 | GLM-5.3 |
| Mystery Game Puzzles (Epoch AI run) | 34% ↗epoch run | 33% ↗epoch run | +1 pt | DeepSeek V4 Flash |
| NL2Repo | 54.2% ↗ | 58% ↗ | -3.8 pt | GLM-5.3 |
| OTIS Mock AIME 2024-2025 (Epoch AI run) | 94.4% ↗epoch run | 91.1% ↗epoch run | +3.3 pt | DeepSeek V4 Flash |
| SimpleBench (SimpleBench) | 61.1% ↗ | 66.2% ↗ | -5.1 pt | GLM-5.3 |
| SimpleQA Verified | 33.6% ↗epoch run | 41% ↗epoch run | -7.4 pt | GLM-5.3 |
| Terminal-Bench 4.0 (Vals AI) | 18.7% ↗ | 38.9% ↗ | -20.2 pt | GLM-5.3 |
| WeirdML (Håvard Tveit Ihle) | 63% ↗ | 75.4% ↗ | -12.4 pt | GLM-5.3 |
| Only DeepSeek V4 Flash reports · 6 | ||||
| AutomationBench Public | 25.1% ↗ | not reported | — | — |
| Cybergym | 76.7% ↗ | not reported | — | — |
| DeepSWE | 54.4% ↗ | not reported | — | — |
| Terminal Bench 2.1 | 82.7% ↗ | not reported | — | — |
| Toolathlon-Verified | 70.3% ↗ | not reported | — | — |
| Toolathlon-Verified (HKUST) | 70.7% ↗ | not reported | — | — |
| Only GLM-5.3 reports · 21 | ||||
| Agents' Last Exam (ALE-CLI) | not reported | 28.5% ↗ | — | — |
| APEX-Agents (Mercor) | not reported | 56.6% ↗ | — | — |
| AutomationBench v1.0.6 | not reported | 48.2% ↗ | — | — |
| CyberGym | not reported | 84.5% ↗ | — | — |
| DeepSWE v1.1 | not reported | 66.9% ↗ | — | — |
| ExploitBench | not reported | 54.4% ↗ | — | — |
| ExploitGym 2h | not reported | 105 ↗ | — | — |
| ExploitGym 6h | not reported | 130 ↗ | — | — |
| FrontierSWE | not reported | 78.1% ↗ | — | — |
| FrontierSWE V2 (Proximal Labs) | not reported | 30.2% ↗ | — | — |
| GDPval-AA v2 | not reported | 1769 ↗ | — | — |
| HLE (with tools) | not reported | 62.5% ↗ | — | — |
| LMArena Agent (LMArena) | not reported | 0.0236 ↗ | — | — |
| PostTrainBench | not reported | 39.8% ↗ | — | — |
| ProgramBench (Almost Solved) | not reported | 19% ↗ | — | — |
| SWE-Marathon v1.1 | not reported | 42.5% ↗ | — | — |
| Terminal-Bench 2.1 | not reported | 88.2% ↗ | — | — |
| Terminal-Bench 3.0 | not reported | 28.3% ↗ | — | — |
| Toolathlon Verified | not reported | 73% ↗ | — | — |
| Vending-Bench 2 (Andon Labs) | not reported | 8163.61 ↗ | — | — |
| Z.ai Code Bench (max effort) | not reported | 34.5% ↗ | — | — |
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-5.3 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-5.3 | 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 | 2026-08-18 | — |
| Cited benchmark scores | 28 | 57 | — |
Reliability
Provider status
Live from /statusMore matchups
DeepSeek V4 Flash vs …
Models sharing the most benchmarksMore matchups
GLM-5.3 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-5.3 on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.