Z.ai · Chat and reasoning · 321B parameters · 18B active · mit
GLM-5.3 Flash
Needs 394 GB at FP8 / INT8. The smallest real configuration today is 2× MI325X at Cyfuture AI, $6.21 per hour, about $4,536 a month at 730 hours.
Parameters321B18B active per token
Context1M tokensmaximum published
Checkpoint328 GBFP8, as published
Licensemitcheck before commercial use
ReleasedAug 25, 2026documented date
Smallest configuration · FP8 / INT8
2× MI325X512 GB VRAM · 77% used
- Per hour
- $6.21
- Per month
- $4,536
- Cheapest at
Cyfuture AI
What is documented about GLM-5.3 Flash
news checked 2026-10-27Released 2026-08-25 under MIT; 320B parameters, 18B active, FP8 checkpoint of 328 GB read on Hugging Face on 2026-10-27. Existence corroborated by third-party benchmarks and vLLM tickets in the 2026-10-25 report.
- GPUBank global report, 2026-10-25, §3.4 (internal, sourced)
- https://huggingface.co/zai-org/GLM-5.3-Flash
- https://huggingface.co/zai-org/GLM-5.3-Flash/raw/main/config.json
- https://huggingface.co/api/models/zai-org/GLM-5.3-Flash?blobs=true
Memory · 2 precisions · 6 context lengthsHow much VRAM it needs
Weights plus runtime overhead, then the KV cache that grows with the context you serve.
| Precision | Weights | Weights + headroom | 4k context | 32k context | 128k context | 256k context | 1M context |
|---|---|---|---|---|---|---|---|
| FP8published checkpoint | 328 GB | 394 GB | 395 GB | 396 GB | 400 GB | 407 GB | 443 GB |
| INT4 / MXFP4estimated | 193 GB | 232 GB | 232 GB | 233 GB | 238 GB | 244 GB | 280 GB |
| KV cacheone sequence, FP16 cache, included above | — | — | 1 GB | 2 GB | 7 GB | 13 GB | 49 GB |
Cross-checked on Hugging Face Oct 28, 2026: 321.3B parameters in the safetensors headers, 328.3 GB of weight files, attention layout from config.json (45 layers, latent rank 512). Method: weights as published at the native precision, verified on Hugging Face 2026-10-28; other precisions estimated at 1 (8-bit) and 0.6 (4-bit) bytes per parameter; × 1.2 runtime headroom; KV cache = 1024 bytes per token per layer × 45 full-attention layers (FP16 cache, one sequence).config.json ↗
Smallest configuration per GPU · 110 GPUs · cheapest eligible on-demand offer per GPU count, every provider on fileWhat it costs to run, GPU by GPU
The fewest GPUs of one type whose total memory fits the model at short context, priced from the cheapest eligible on-demand offer for exactly that count (45-day window, newest reading per provider), with the next providers beside it. Monthly = 730 hours.
| GPU | Smallest config | VRAM used | $ / hour | $ / month | Cheapest at |
|---|---|---|---|---|---|
| MI325XBest fit | 2× 256 GB512 GB total | 77% | $6.21 | $4,536 | Cyfuture AInot scored · read Nov 1 · receipt |
| RTX PRO 6000 | 8× 96 GB768 GB total | 51% | $7.78 | $5,676 |  UpCloudTrust 75 · read Nov 1 · receipt · also Vast.ai $10.14, Lium $12.80 |
| Gaudi 2 | 8× 96 GB768 GB total | 51% | $8.27 | $6,036 | Cyfuture AInot scored · read Nov 1 · receipt |
| A100 80GB | 8× 80 GB640 GB total | 62% | $9.82 | $7,166 | Edgevananot scored · read Nov 1 · receipt · also GPU.ai $10.00, io.net $10.50 |
| H200 NVL | 4× 141 GB564 GB total | 70% | $10.67 | $7,788 |  Vast.aiTrust 78 · read Nov 1 · receipt · also IONOS $13.04, Geodd $13.56 |
| H200 | 4× 141 GB564 GB total | 70% | $11.05 | $8,070 | Vast.aiTrust 78 · read Nov 1 · receipt · also Theta EdgeCloud $14.76, Canopy Wave $16.00 |
| MI300X | 4× 192 GB768 GB total | 51% | $11.31 | $8,257 | Cyfuture AInot scored · read Nov 1 · receipt · also Hot Aisle $11.96, AtmosCompute $13.38 |
| B300 | 2× 288 GB576 GB total | 68% | $11.65 | $8,506 | UpCloudTrust 75 · read Nov 1 · receipt · also FPT AI Factory $13.98, Vast.ai $17.00 |
| H100 | 8× 80 GB640 GB total | 62% | $13.67 | $9,982 |  GcoreTrust 80 · read Nov 1 · receipt · also Lium $14.36, Airon AI $15.04 |
| B200 | 4× 180 GB720 GB total | 55% | $16.46 | $12,013 | UpCloudTrust 75 · read Nov 1 · receipt · also Deep Infra $19.56, Lambda $27.16 |
| MI355X | 8× 288 GB2304 GB total | 17% | $23.60 | $17,228 |  TensorWaveTrust 64 · read Nov 1 · receipt |
| GB300 | 2× 288 GB576 GB total | 68% | $26.11 | $19,060 |  VerdaTrust 81 · read Nov 1 · receipt |
| B200 CC | 4× 180 GB720 GB total | 55% | $36.19 | $26,419 | VerdaTrust 81 · read Nov 1 · receipt |
| B300 CC | 4× 288 GB1152 GB total | 34% | $46.31 | $33,806 | VerdaTrust 81 · read Nov 1 · receipt |
| A800 80GB PCIe | 8× 80 GB640 GB total | 62% | $50.80 | $37,088 |  Volcano Enginenot scored · read Nov 1 · receipt |
| H100 NVL | 8× 94 GB752 GB total | 52% | $51.79 | $37,809 |  Open Telekom Cloudnot scored · read Nov 1 · receipt · also Cloud.ru $52.63 |
Throughput estimates · internal study · 2026-10-04Tokens per second, and what a million costs
Chat and reasoning. Speed estimated from memory bandwidth and model size, never measured. The cost per 1M output tokens is the study's own estimate at the prices it read on 2026-10-04; it is not recomputed here.
| GPU | Cards | Precision | tok/s per user | per 1M output tokens · study estimate |
|---|---|---|---|---|
| MI300X | 2 | FP8 | 60.7 | $1.07 |
| H200 | 4 | FP8 | 69.1 | $1.82 |
| B200 | 4 | FP8 | 92.9 | $1.94 |
| A100 80GB | 8 | FP8 | 46.8 | $2.13 |
| RTX PRO 6000 | 4 | FP8 | 39.4 | $2.39 |
| H100 SXM | 8 | FP8 | 57.9 | $2.51 |
| H100 PCIe | 8 | FP8 | 34.1 | $4.03 |
| L40S | 8 | FP8 | 27.2 | $9.49 |
cards, precision and speed estimated from memory bandwidth and model size (internal study); cost recomputed by the site from verified prices · cheapest per output at study prices: MI300XWhy estimates, and their limits →
GLM-5.3 Flash · related