The short URL: https://ask.yugatech.com/go/q257

Z.ai has launched GLM-5.3-Flash, the first natively multimodal model in its GLM-5 series. Unlike a text model with a vision feature bolted on, this one is built from the ground up to handle text, images, and video in a single stack, according to the company.

GLM 5. Flash model scaled | Z.ai launches GLM-5.3-Flash, a multimodal model with a 1M-token context window

GLM-5.3-Flash uses a mixture-of-experts design, meaning only a fraction of its neural network activates per request. The full model spans 320 billion parameters, but only about 18 billion are active at any one time, which keeps cost and response time down. Z.ai says the model was trained on 30 trillion tokens, the text fragments AI systems learn from.

A long context window. The model reads up to 1 million tokens of context at once, roughly 1,600 pages of text, per Z.ai. That lets it hold an entire long document or a long conversation in memory instead of forgetting earlier parts. Vision also sits inside the coding loop, which Z.ai says lets the model judge its own work visually while writing code.

GLM 5. Flash model 1 scaled | Z.ai launches GLM-5.3-Flash, a multimodal model with a 1M-token context window

Benchmarks. Z.ai’s self-reported scores show big jumps over the previous GLM-5.2 on agentic coding tasks, which measure how well an AI can work through multi-step problems rather than answer a single prompt. DeepSWE rose to 63.4 from 46.2, and AutomationBench to 48.8 from 26.2. On several agent benchmarks the model approaches the closed-source Opus 4.8. These are vendor numbers and have not been independently verified.

API access starts at USD 0.15 ( PHP 9.26 ) per million input tokens, with cached input at USD 0.03 ( PHP 1.85 ) and output at USD 0.50 ( PHP 30.86 ). A 50% promo runs through September 9, 2026. That makes Flash roughly 10x cheaper on input than the text-only GLM-5.3 flagship, which lists at USD 1.40 per million tokens.

GLM 5. Flash model 2 scaled | Z.ai launches GLM-5.3-Flash, a multimodal model with a 1M-token context window

Open weights and availability. GLM-5.3-Flash is released under the MIT license, which permits commercial use, and the weights are available through Z.ai’s repository. The model is served through Z.ai’s API as well as DeepInfra, Novita, and OpenRouter. Z.ai first tested the model under the stealth name ox-alpha.

Source

GLM-5.3-Flash specs:
Model: GLM-5.3-Flash, first natively multimodal GLM-5
Architecture: 320B total / 18B active, mixture of experts, 45 layers
Context: 1M tokens input, 131K tokens output
Modalities: text, image, video input
Training: 30 trillion tokens
License: MIT (commercial use allowed)
Pricing: USD 0.15 / 0.03 / 0.50 per 1M tokens (input / cached / output); 50% promo until September 9, 2026
Providers: Z.ai API, DeepInfra, Novita, OpenRouter