GLM-5.3-FlashX

Z.AI's GLM-5.3-FlashX is a speed-optimized native multimodal large model from Z.AI built on an efficient sparse architecture with 320B total parameters and 18B activated parameters. It boosts generation speeds to up to 200 tokens per second, delivering approximately a 5x acceleration over GLM-5.3-Flash while maintaining original intelligence standards. The model supports ultra-long 1M-token context windows and comprehensive native understanding of videos and files. GLM-5.3-FlashX includes interleaved tool-calling reasoning capabilities designed for high-frequency interactive programming and agent workflows. The model became accessible simultaneously via API and Web Chat interfaces on the B.AI platform on September 21, 2026, allowing developers to directly invoke it. Creator attribution is preserved to Z.AI, with B.AI noted only as the hosting platform where availability was announced.

Command Code confirmed GLM-5.3-FlashX went live on September 18, 2026, with the model ID z-ai/glm-5.3-flashx across all plans and API access. The integration specs call out 200 TPS, a 1M-token context window, and native multimodal support for text, image, and video inputs — matching Zhipu's positioning of FlashX as a high-speed variant of GLM-5.3-Flash. Developers can invoke it via CLI or switch to it mid-session without losing context. Intelligence and Coding Index scores referenced for the model are attributed to that benchmark provider rather than newly run by the integrator, and Command Code's cost-per-task examples are illustrative rather than vendor-published guarantees. The page snapshots the FlashX model listing for developer reference but functions as downstream consumer coverage rather than a primary Zhipu release note. Its main value is confirming the exact variant's spec sheet and identifier as published to third-party tooling.

TechFlow's September 18, 2026 newsletter reports Zhipu's official launch of GLM-5.3-FlashX, confirming the model was previously previewed to global developers under the codename "Ox Alpha." Zhipu stated the 200 tokens/s speed was achieved through increased infrastructure investment and inference optimization on a computing cluster backed by 100,000 domestically produced chips, with the FlashX API going live under the identifier "GLM-5.3-FlashX." The launch framing positions FlashX as combining intelligence, price, and speed simultaneously for the first time within the GLM-5 series. The article notes rapid prior adoption of the underlying GLM-5.3-Flash among global developers drove Zhipu to scale inference capacity, making FlashX a direct response to enterprise demand for higher-throughput, lower-latency access to the same model weights.

An aggregator news feed entry dated September 18, 2026, lists Z.AI's GLM-5.3-FlashX as the high-speed variant of GLM-5.3-Flash, a native multimodal model delivering inference speeds of up to 200 tokens per second. It is built on the same hybrid sparse and linear attention architecture as the base Flash variant. This corroborates the model variant's release timeline and core technical attributes. The same feed situates GLM-5.3-FlashX within Z.AI's broader model family alongside GLM-5.3-Prime and the GLM Flash series, confirming it as a distinct speed-optimized variant. The entry traces its origin to an OpenRouter listing, serving as secondary corroboration rather than a primary Z.AI source. Creator attribution to Z.AI is preserved throughout the aggregator's description.

GLM-5.3-FlashX uses the same weights as GLM-5.3-Flash — a mixture-of-experts architecture with roughly 320 billion total parameters and about 18 billion active per token. It is natively multimodal, accepting text, images, and video as input while producing text output, and supports a 1M-token context window with a 128K maximum output length. The Z.ai API model identifier is glm-5.3-flashx, confirming exact-variant availability on the official vendor. No new benchmark numbers were published for FlashX because the weights did not change; the up-to-200 tokens/s speed gain comes entirely from serving-layer optimization on Zhipu's domestic-chip inference cluster. This makes FlashX a time-to-first-answer improvement rather than a smarter model, and aggregator catalogs at launch had not yet listed a FlashX slug alongside the existing Flash entry. The guidance notes that developers should copy the exact identifier from the vendor catalog rather than assuming sibling slugs will resolve.

Zhipu released GLM-5.3-FlashX on September 18, 2026, as a high-throughput serving variant of GLM-5.3-Flash, achieving up to 200 tokens per second. The model is framed as a speed-focused optimization that keeps Flash's intelligence and price advantages while delivering roughly five times Flash's output speed. The launch also brought the FlashX API live, giving enterprise developers direct access to the faster variant alongside the existing Flash endpoint. FlashX runs on Zhipu's scaled domestic-chip inference cluster, reflecting continued investment in infrastructure and inference optimization to meet rising developer demand. The release is positioned as a serving-layer speed upgrade rather than a new capability, since the underlying model weights remain unchanged from GLM-5.3-Flash. The launch is part of Zhipu's broader push to balance intelligence, price, and speed simultaneously for the first time at this scale.

Z.ai's official developer documentation introduces GLM-5.3-FlashX as the high-speed variant of GLM-5.3-Flash, achieving inference speeds of up to 200 tokens/s on the same multimodal architecture. The model shares Flash's 320B total / 18B active parameters, hybrid sparse-plus-linear attention, 1M-token context window, and 128K maximum output, with API code "glm-5.3-flashx" available via the Chat Completion API. No new benchmarks were published because weights remain unchanged from Flash. The documentation positions FlashX alongside Flash on the GLM Coding Plan, emphasizing native multimodal visual coding, professional workflow support (Office, financial research, document processing), and recommended sampling settings of temperature 1, top_p 0.95, and reasoning_effort max. The FlashX-specific speed gain is delivered through inference optimization on Zhipu's domestic-chip infrastructure rather than capability changes, making it a serving-layer upgrade targeted at high-throughput, low-latency enterprise usage.