Step 5 Preview vs GLM 5.3 Flash
Step 5 Preview leads llmrun Score 128.0 to 124.1 (ranges overlap); neither fits 24 GB.
Model A · StepFun
Step 5 Preview604.3B · 27B active · 1M context · Other · Sep 2026
Model B · Z.ai
GLM 5.3 Flash321.3B · 18.3B active · 1M context · MIT · Aug 2026
Choose Step 5 Preview if you want…
Higher llmrun Score
128.0 vs 124.1, ranges overlap
Stronger at science
134.2 vs 123.5
Ahead on the benchmarks
3 of 3 shared boards
Also does chat
GLM 5.3 Flash doesn't list it
Choose GLM 5.3 Flash if you want…
Lighter on memory
194.6 vs 398.9 GB at Q4_K_M
Also does vision
Step 5 Preview doesn't list it
More permissive license
MIT vs Other
On your hardware
Each model at its default quant (Q4_K_M), a smaller one when it must. Estimates, ±20%.
8 GB
RTX 4060 · 8 GB tier
Step 5 Preview
Doesn't fit
smallest quant needs 191.4 GB
GLM 5.3 Flash
Doesn't fit
smallest quant needs 100.7 GB
12 GB
RTX 3060 12GB · 12 GB tier
Step 5 Preview
Doesn't fit
smallest quant needs 191.4 GB
GLM 5.3 Flash
Doesn't fit
smallest quant needs 100.7 GB
16 GB
RTX 5060 Ti 16GB · 16 GB tier
Step 5 Preview
Doesn't fit
smallest quant needs 191.4 GB
GLM 5.3 Flash
Doesn't fit
smallest quant needs 100.7 GB
24 GBreference
RTX 4090 · 24 GB tier
Step 5 Preview
Doesn't fit
smallest quant needs 191.4 GB
GLM 5.3 Flash
Doesn't fit
smallest quant needs 100.7 GB
32 GB
RTX 5090 · 32 GB tier
Step 5 Preview
Doesn't fit
smallest quant needs 191.4 GB
GLM 5.3 Flash
Doesn't fit
smallest quant needs 100.7 GB
48 GB
RTX A6000 · 48 GB tier
Step 5 Preview
Doesn't fit
smallest quant needs 191.4 GB
GLM 5.3 Flash
Doesn't fit
smallest quant needs 100.7 GB
Mac 32 GB
Mac Mini M4 · Macs
Step 5 Preview
Doesn't fit
smallest quant needs 191.4 GB
GLM 5.3 Flash
Doesn't fit
smallest quant needs 100.7 GB
Mac 64 GB
Mac Studio M4 Max · Macs
Step 5 Preview
Doesn't fit
smallest quant needs 191.4 GB
GLM 5.3 Flash
Doesn't fit
smallest quant needs 100.7 GB
Mac 128 GB
Mac Studio M4 Max · Macs
Step 5 Preview
Doesn't fit
smallest quant needs 191.4 GB
GLM 5.3 Flash
IQ2_S · ~41 tok/s
112.8 GB · up to 16Kfits
Reference GPU: RTX 4090. Tinted cell = the better pick on that machine (fits, more headroom, or at least 10% faster).
Overall
Step 5 Preview
GLM 5.3 Flash
Coding
Step 5 Preview
no data
GLM 5.3 Flash
Agents
Step 5 Preview
no data
GLM 5.3 Flash
Math
Step 5 Preview
no data
GLM 5.3 Flash
Science
Step 5 Preview
GLM 5.3 Flash
Reasoning
Step 5 Preview
no data
GLM 5.3 Flash
Step 5 Preview · #8 of 101 open models · #6 of 98 that fit ≤512 GB
GLM 5.3 Flash · #11 of 101 open models · #6 of 86 that fit ≤256 GB
Benchmarks
3 shared boards: Step 5 Preview leads 3, GLM 5.3 Flash leads 0 · Benchmarks imported Oct 2026
Step 5 Preview
58.9%#4/68 open
Lab-reportedEpoch AI · Sep 2026
GLM 5.3 Flash
51.6%#8/68 open
Lab-reportedEpoch AI · Aug 2026
Step 5 Preview
20.9%#3/68 open
Lab-reportedEpoch AI · Sep 2026
GLM 5.3 Flash
15.4%#8/68 open
Lab-reportedEpoch AI · Aug 2026
Step 5 Preview
42.0%#9/27 open
Lab-reportedEpoch AI · Sep 2026
GLM 5.3 Flash
21.0%#13/27 open
Lab-reportedEpoch AI · Aug 2026
28 more boards (other shared and single-model boards)
Only GLM 5.3 Flash (28)
- GPQA Diamond90.1%
- LMArena Text1476.0
- LiveBench Coding79.0
- LiveBench Reasoning77.6
- AIME 2024/202593.9%
- FrontierMath55.8%
- LMArena WebDev1607.7
- LiveBench Math81.2
- ARC-AGI-265.8%
- ARC-AGI91.0%
- FrontierMath Tier 417.1%
- DeepSWE63.4%
- FrontierCode31.8%
- APEX-Agents52.8%
- Chess Puzzles21.0%
- LiveBench 2026 Reasoning77.6
- Mystery Game Puzzles8.0%
- GDP.pdf14.0%
- LiveBench 2026 Browser84.0
- LiveBench 2026 Data Analysis76.4
- ALE-Bench303.6
- CursorBench36.8%
- FrontierSWE18.1%
- LiveBench 2026 Coding80.6
- LiveBench 2026 Long-Horizon28.4
- Surface Evolver Bench52.5%
- LiveBench 2026 Design77.2
- LiveBench 2026 Research70.1
VRAM by quantization
Memory to load the model and 2K context; below it, the total at 32K and 128K context. Lighter side in bold.
Q2_K
Step 5 Preview
283 GB182.5 GB file
32K 301 GB · 128K 360 GB
GLM 5.3 Flash
138 GB114.2 GB file
32K 161 GB · 128K 234 GB
Q3_K_M
Step 5 Preview
324 GB~294.6 GB file
32K 343 GB · 128K 402 GB
GLM 5.3 Flash
158 GB149.1 GB file
32K 181 GB · 128K 254 GB
Q4_K_M
Step 5 Preview
399 GB~362.6 GB file
32K 417 GB · 128K 476 GB
GLM 5.3 Flash
195 GB188.7 GB file
32K 217 GB · 128K 290 GB
Q5_K_M
Step 5 Preview
474 GB~430.6 GB file
32K 492 GB · 128K 551 GB
GLM 5.3 Flash
231 GB~228.9 GB file
32K 253 GB · 128K 326 GB
Q6_K
Step 5 Preview
548 GB~498.6 GB file
32K 567 GB · 128K 626 GB
GLM 5.3 Flash
267 GB~265.1 GB file
32K 290 GB · 128K 362 GB
Q8_0
Step 5 Preview
665 GB~604.3 GB file
32K 683 GB · 128K 742 GB
GLM 5.3 Flash
323 GB341.0 GB file
32K 346 GB · 128K 418 GB
6 more quants
IQ2_XXS
Step 5 Preview
183 GB~166.2 GB file
32K 201 GB · 128K 260 GB
GLM 5.3 Flash
90.2 GB101.8 GB file
32K 113 GB · 128K 185 GB
IQ2_S
Step 5 Preview
not listed
GLM 5.3 Flash
102 GB113.6 GB file
32K 125 GB · 128K 197 GB
IQ3_XXS
Step 5 Preview
not listed
GLM 5.3 Flash
126 GB120.4 GB file
32K 149 GB · 128K 221 GB
IQ3_M
Step 5 Preview
299 GB271.6 GB file
32K 318 GB · 128K 377 GB
GLM 5.3 Flash
not listed
IQ4_XS
Step 5 Preview
not listed
GLM 5.3 Flash
175 GB156.8 GB file
32K 197 GB · 128K 270 GB
BF16
Step 5 Preview
1330 GB~1208.7 GB file
32K 1348 GB · 128K 1407 GB
GLM 5.3 Flash
644 GB641.6 GB file
32K 667 GB · 128K 740 GB
Step 5 Preview
Ollama · Q4_K_M
ollama run hf.co/smalinin/Step-5-Preview-GGUFGLM 5.3 Flash
Ollama · Q4_K_M
ollama run hf.co/batiai/GLM-5.3-Flash-GGUF:Q4_K_M| Spec | Step 5 Preview | GLM 5.3 Flash |
|---|---|---|
| Publisher | StepFun | Z.ai |
| Parameters | 604.3B | 321.3B |
| Active parameters | 27B per token | 18B per token |
| Architecture | Step5ForConditionalGeneration | Glm5NextForConditionalGeneration |
| Context | 1M tokens | 1M tokens |
| Modalities | Chat, Function-calling | Vision, Function-calling |
| License | Other | MIT |
| Released | Sep 2026 | Aug 2026 |
Frequently Asked Questions
Which is better, Step 5 Preview or GLM 5.3 Flash?
Step 5 Preview leads llmrun Score 128.0 to 124.1 (ranges overlap); neither fits 24 GB. Choose Step 5 Preview for higher llmrun Score and stronger at science; choose GLM 5.3 Flash for lighter on memory and also does vision.
Which needs less VRAM, Step 5 Preview or GLM 5.3 Flash?
At Q4_K_M, Step 5 Preview requires 398.9 GB while GLM 5.3 Flash needs 194.6 GB, making GLM 5.3 Flash the easier model to run on your local setup.
Can I run Step 5 Preview or GLM 5.3 Flash on a 16 GB GPU?
Step 5 Preview does not fit (smallest quant needs 191.4 GB); GLM 5.3 Flash does not fit (smallest quant needs 100.7 GB).
Which has a longer context window, Step 5 Preview or GLM 5.3 Flash?
Step 5 Preview supports 1,048,576 tokens and GLM 5.3 Flash supports 1,048,576 tokens.