GLM 5.3 Flash

GLM-5.3-Flash was released by Z.ai on August 26, 2026, under the MIT license with weights published on Hugging Face the same day. According to the page, it is a 320-billion-parameter mixture-of-experts model that activates only 18 billion parameters per token, making it the first natively multimodal model in the GLM-5 The article documents Z.ai's launch pricing of $0.15 per million input tokens and $0.50 per million output tokens, roughly a tenth of GLM 5.3's $1.40 and $4.40 pricing, and contrasts the model's benchmark posture against the still-unreleased GLM 5.3 (744B) weights. It highlights that while Flash posts competitive numbe

Before its identity was revealed, the model appeared on OpenRouter and OpenCode on August 20, 2026, listed only as "Ox Alpha" with no owner, no model card, a 1M-token context window, and text/image/video input. Within six days it became the most-used model on OpenRouter, processing roughly 23 trillion tokens — about 2. Weights landed on Hugging Face under an MIT license the same evening as the reveal, and Zhipu's Hong Kong shares closed more than 12% higher the next session, roughly 10x the January IPO price according to the article. Stripe had agreed to acquire OpenRouter the day before the model appeared, and Stripe CEO Patrick Col

Ampere's comparison covers GLM 5.3 Flash's positioning relative to the flagship GLM 5.3, reporting Artificial Analysis Intelligence Index scores of 57 for Flash and 60 for GLM 5.3, alongside total/active parameter counts of 320B/18B and 753B/40B respectively. The page documents a counterintuitive speed finding: GLM 5.3 Pricing data confirms $0.15 per million input and $0.50 per million output for Flash against $1.40 and $4.40 for the flagship, with Flash also offering native image input and visual coding capabilities that GLM 5.3 lacks. The article recommends Flash as the better value for most high-volume applications and production

DataCamp's coverage confirms GLM-5.3-Flash as Z.ai's cost-optimised sibling to the GLM-5.3 flagship, codenamed "Ox Alpha" during a stealth preview period before public reveal. The model uses a 320B-A18B mixture-of-experts configuration, supports a 1M-token context window, and is natively multimodal, distinguishing it f The article reports Z.ai's own benchmark of 84.3 on Terminal-Bench 2.1, placing GLM-5.3-Flash within striking distance of Claude Opus 4.8 (85.0) and GPT-5.6 Terra (87.4), and cites a blended API price of roughly $0.10 per million tokens compared to about $0.90 for GLM-5.3. It notes the practical deployment constraint t

Z.ai has introduced GLM-5.3-Flash, the first natively multimodal model in the GLM-5 series, released on August 26, 2026 under the MIT License with weights on Hugging Face. The model uses a mixture-of-experts design with 320B total parameters and 18B active per token, paired with a 1M-token context window and a hybrid s On benchmarks, GLM-5.3-Flash scores 57 on the Artificial Analysis Intelligence Index v4.1.1 at roughly $0.045 per task and approaches Claude Opus 4.8 on coding and agentic evaluations while outperforming GLM-5.2 across real-world workloads at about one-tenth the price. Standard Z.ai API pricing is $0.15 per 1M input to

GLM-5.3-Flash is a 320B-parameter mixture-of-experts model with 18B active parameters per token, runs natively in FP8, supports a 1,048,576-token context window, and is the first natively multimodal model in the GLM-5 series. The technical deep dive describes the architecture as a hybrid sparse and linear attention des API pricing is confirmed at $0.15 per million input tokens and $0.50 per million output tokens, with the launch table showing Flash matching or beating the GLM 5.3 flagship on several agent benchmarks. The article highlights that GLM 5.3's promised 744B-parameter open weights have not shipped as of launch day, leaving

Together AI's model page lists GLM-5.3-Flash under the model key zai-org/GLM-5.3-Flash, framing it as frontier coding and multimodal intelligence engineered for flash-level inference cost. It restates the 320B-total/18B-active MoE with a hybrid linear-plus-sparse attention design and a lightweight indexer that cuts att The Together AI entry emphasizes native multimodal input that brings vision into the coding loop, letting the model inspect its own rendered outputs and iteratively refine them, and read professional documents, spreadsheets, dashboards, and interfaces. It reiterates that GLM-5.3-Flash was previewed anonymously as Ox Al

Z.ai's official release-notes index lists GLM-5.3-Flash under a 2026-08-26 entry, describing it as natively multimodal so the model can observe interfaces, rendered results, and interaction feedback to close a loop across code, browsers, and GUIs. The entry confirms the efficient hybrid architecture combining linear an Surrounding entries provide release-lineage context: GLM-5.3 on 2026-08-18 reports a 50% gain over GLM-5.2 on Z.ai Code Bench with SOTA open-source results on Terminal Bench 3.0 plus cybersecurity capabilities matching Mythos 5; GLM-5.2 on 2026-06-16 added 1M lossless context; and GLM-5.1 on 2026-04-07 introduced long-