Z.ai
GLM 5.3 Flash
An A-tier efficiency build: the full 16-prompt visual set delivered at a fraction of frontier cost.
Verified model facts
Identity, limits, and pricing
- Status
- Current
- API model ID
glm-5.3-flash- Context
- 1M
- Max output
- 131K
- License
- Not publicly verified
- Access route
- Not publicly verified
- Modalities
- Not publicly verified
- Hardware
- Not publicly verified
- API price / 1M tokens
- $0.15 input · $0.03 cached input · $0.5 output
Z.ai list price; launch-capacity resellers sold at roughly half price through September 9, 2026.
Same-prompt visual tests
Inspect the actual runs
Open the generated scene to test it, or compare that prompt with another model.
City Scroll Journey
Cinematic scroll journey testing AI-generated scene continuity, scroll-scrubbed camera motion, and art-directed landing craft.
Ember Glider
Sunset gliding journey testing flight energy management, checkpoint flow, and atmospheric scene craft.
Helm's Deep
Fortress siege scene testing scale, lighting, architecture, and cinematic atmosphere.
Hogwarts Broom Flight Simulator
Broom-flight scene testing depth, motion cues, castle scale, and fantasy mood.
Jabberwock
Dark fantasy encounter testing creature design, forest mood, and narrative staging.
Low-Poly Tower Defense
Diorama tower defense testing economy balance, wave design, placement rules, and combat readability.
Low Poly World
Stylized island build testing composition, color, and low-poly worldbuilding.
Mechanical Watch Simulator
Interactive watch movement testing mechanical legibility, accurate relative motion, and real-time 3D controls.
Neon Drift
Synthwave time-trial racing testing drift physics, lap timing, ghost replay, and unlock progression.
Office Life
Workplace vignette testing everyday scene logic, objects, and believable office detail.
Petri Dish
Microscopic ecosystem testing organic forms, scientific clarity, and cellular detail.
Starfall Arena
Neon arena survival testing wave escalation, upgrade builds, particle feedback, and boss design.
Stormwind Trebuchet Simulator
Counterweight siege simulation testing coupled mechanics, trajectory prediction, projectile cameras, and interactive tuning.
Universe Simulator
Cosmic system testing orbital structure, glowing bodies, scale, and simulation readability.
Vice City
Neon coastal city testing vehicles, architecture, atmosphere, and dense urban layout.
Yingzao Fashi Assembly
Timber assembly scene testing structure, joinery, construction order, and material clarity.
Official benchmark profile
GLM-5.3-Flash efficiency-tier coding and agent results.
Z.ai positions GLM-5.3-Flash as the efficiency variant of GLM-5.3: a 320B-parameter open-weight model (18B activated) with hybrid sparse-plus-linear attention, native text, image, and video input, and roughly one-tenth the list price of frontier APIs. Z.ai reports it approaches Claude Opus 4.8 across six coding and agentic benchmarks.
DeepSWE v1.1
63.4%AutomationBench
48.8%Z.ai Code Bench v1.0
29.0%Full official benchmark table6 rows with source settings and peer charts
| Benchmark | Area | Score | Setting / comparison |
|---|---|---|---|
| DeepSWE v1.1 | Coding | 63.4% | Z.ai launch result; GLM-5.2 scored 46.2% in the same comparison. |
| AutomationBench | Agentic work | 48.8% | Z.ai launch result; GLM-5.2 scored 26.2% in the same comparison. |
| Z.ai Code Bench v1.0 | Coding | 29.0% | Claude Code 2.1.207 harness at maximum reasoning effort; Claude Opus 4.8 scored 29.5% on the same harness. |
| Artificial Analysis Intelligence Index v4.1.1 | Composite intelligence | 57 | Z.ai-reported score at a discounted $0.045 per task, which Z.ai says pushes the cost-quality Pareto frontier. |
| GPQA Diamond | Academic reasoning | 87.5% | Highest provider-reported OpenRouter result (GMICloud); other providers reported 84.3–84.5%. |
| TAU-Bench | Tool use | 80.7% | Highest provider-reported OpenRouter result (NovitaAI); the Z.ai direct endpoint reported 77.3%. |
Editorial verdict
Where this model fits
GLM 5.3 Flash is Z.ai's efficiency variant of GLM 5.3: a 320B-parameter open-weight model (18B activated) with native text, image, and video input at roughly a tenth of frontier list price. Independent builder agents completed all sixteen Superbash visual prompts in one series, every artifact validating and running, and Z.ai reports coding results approaching Claude Opus 4.8.
Best for
Watch out
Why it is ranked here
Evidence
Superbash editorial model ranking
Takeaway: GLM 5.3 Flash is currently placed in Tier A.
The August 2026 editorial roster places GLM 5.3 Flash at rank 8.
Open source →GLM-5.3-Flash
Takeaway: Use GLM 5.3 Flash for cost-conscious coding work, then review the output.
The provider page documents the hybrid sparse-plus-linear architecture, multimodal input, and Coding Plan availability with three times the GLM-5.3 quota.
Open source →Related guides