Z.ai has launched GLM-5.3-Flash, the first natively multimodal model in its GLM-5 family, as a lower-cost option for coding, agent tasks, and visual reasoning. The mixture-of-experts model has 320 billion total parameters and 18 billion active parameters, down from 32 billion active parameters in GLM-4.5. Z.ai says it outperforms GLM-5.2 across reported coding and agentic tests at one-tenth the price while approaching Claude Opus 4.8 on its internal coding benchmark.
Trained on a 30-trillion-token multimodal corpus, GLM-5.3-Flash combines linear attention for local dependencies with sparse attention for relevant global context. At context lengths reaching one million tokens, IndexPool compresses groups of indexer key vectors to limit latency and memory use. Z.ai reports three times less attention compute and a 4.4-fold reduction in KV cache size compared with GLM-5.3.
Introducing GLM-5.3-Flash
— Z.ai (@Zai_org) August 26, 2026
- Leading capabilities at a highly competitive price
- Natively multimodal with a 1M-token context window
- A 320B-A18B model released under the MIT License
- Previously previewed as Ox Alpha, running entirely on Chinese AI chips
Blog:… pic.twitter.com/KOCG4dkay3
On Artificial Analysis Intelligence Index v4.1.1, the model scored 57 at a discounted cost of $0.045 per task, according to Z.ai. It also reached 63.4 on DeepSWE v1.1 against GLM-5.2's 46.2, and 48.8 on AutomationBench against 26.2. Because the evaluations use different harnesses, context limits and generation settings, comparisons depend on each test setup.
Visual reasoning is central to the release. Z.ai trained the model to inspect rendered interfaces, gameplay and 3D output, then assess and revise its work from visual feedback. The same approach covers documents, spreadsheets, presentations, dashboards and meeting materials, allowing the model to reason across text, images and structure.
GLM-5.3 Flash, aka Ox Alpha, has officially arrived. The announcement and benchmarks are up already, and the Pareto Frontier has been redrawn. https://t.co/jVij8PJ5xc pic.twitter.com/Z0Ug3nmnmL
— Andrew Curran (@AndrewCurran_) August 26, 2026
Before launch, GLM-5.3-Flash appeared anonymously as ox-alpha on OpenCode and OpenRouter. Z.ai says it became the most popular model of the week on those services, with traffic served on Chinese AI chips. The company built an SGLang-based stack that separates encoding, prefill and decoding, reporting a threefold gain in end-to-end serving performance across tens of thousands of domestic accelerators.
GLM-5.3-Flash is now available to all GLM Coding Plan users with three times the usable quota of GLM-5.3. Its multimodal capabilities are offered in ZCode through Browser Use and Computer Use. The weights are available on Hugging Face, and local deployment supports SGLang, vLLM and TokenSpeed.
