GLM-5.3-Flash
Zhipu AI · Undisclosed (distilled lightweight)
GLM-5.3 capability at ¥3
Released 16 October 2026 as the lightweight GLM-5.3: documents 96, perception 95 and agents 94 sit only one to two points below the flagship, at a third of its price (¥3 versus ¥9), with an ecosystem score of 98 — the top band here. Composite 93.7 ranks #7 overall, the first genuinely usable option for “domestic-chip plus tight budget”. The trade-offs are explicit: video 90 and generation 86, so keep long video and creative work away from it.
Last updated Oct 18, 2026Editorial scoring
Bottom line
Eight-dimension breakdown
editorial mapping · 0–100- GLM-5.3-Flash93.7
Weights: perception 20 / documents 18 / charts 14 / video 14 / agents 14 / generation 6 / Chinese 6 / ecosystem 8. Definitions and sources on the methodology page.
Public benchmarks
as reported, not converted| Benchmark / metric | Value | Note |
|---|---|---|
| SuperCLUE-VLM 2026-10 | 88.6 | Vendor figure; up from GLM-5 |
| 输出价格 | ¥3 | One third of GLM-5.3 |
| 上下文 | 200K | Same as GLM-5 |
Public scores come from different versions and prompts and are not interchangeable with our composite; they are one evidence input.
Scenario mapping
five stress tests, one scale| Scenario | Score | Observation |
|---|---|---|
| Chinese receipt OCR | 95 · Stable on clean receipts and official forms | |
| Chart data extraction | 92 · Reliable on standard charts; average on stacked ones | |
| 30-min video QA | 88 · Long video remains the weak spot | |
| Mobile GUI control | 93 · Well-formed tool calls; above-average recovery | |
| Chinese cultural context | 94 · Solid on official register |
Strengths and limitations
Strengths
- At ¥3/1M output tokens it is the cheapest in its capability band, Ascend images and on-prem licence included.
- Documents 96, perception 95 and agents 94 sit 1–2 points below the flagship — enough for document batches and tool use.
- Ecosystem 98 — official SDK, transparent rate limits, domestic-accelerator support — top band on the board.
Limitations
- Video 90 and generation 86: neither long-video QA nor creative generation is its scenario.
- Parameters and training details are undisclosed; the distillation source cannot be verified.
- Grade B: vendor figures only, with third-party reproduction still pending.
Price and real cost
List price
¥3/M tokens
≈ ¥3 / 1M output tokens. For 1,000 images at 1,024px, ~1,100 image tokens each and 300 output tokens, the monthly cost is about ¥1.8 (assuming input at 25% of output price).
Use your volumeAbout GLM-5.3-Flash
Is GLM-5.3-Flash open source?
Licence: Commercial + on-prem licence (Ascend images included). No public weights; access is via vendor API or an on-prem licence.
How much does GLM-5.3-Flash cost?
≈ ¥3 / 1M output tokens. It ranks #7 with a composite of 93.7. Real bills depend on resolution, tiling, caching and volume discounts.
What is GLM-5.3-Flash best for?
Domestic-chip projects on a tight budget; Document batches and receipt extraction; Lightweight agents and tool orchestration。
Why does GLM-5.3-Flash score 93.7?
The composite weights eight dimensions: perception 20%, documents 18%, charts 14%, video 14%, agents 14%, generation 6%, Chinese 6%, ecosystem 8%. Its strongest dimension is Ecosystem, cost & availability; its weakest is Generation & editing.
Personal ranking · Not an official score
Every ranking, score and comparison on this page is an independent personal ranking, not an official score, and represents no vendor, institution or regulator. Scores are a subjective mapping of public evidence under our own weights — change the weights and the order changes. Disagree, or spotted a factual error? Contact the site owner: jizhi2024@foxmail.com — we check each report and log corrections in the changelog.