GLM 5.3 Flash

Baseten's model library page for GLM-5.3 Fast documents it as a speed-optimized Model API built for real-time workloads, developed by Z AI, version 5.3, with a 753B-A40B MoE size and a custom license. Pricing is listed at $2.10 per 1M input tokens, $0.21 per 1M cache-input tokens, and $6.60 per 1M output tokens — mater The page makes clear GLM-5.3 Fast runs on the same 744B-A40B MoE base as GLM-5.2 and that every capability gain over GLM-5.2 came from scaled post-training, including a Terminal-Bench 3.0 jump from 4.6% to 28.3%. This entry is critical disambiguation: Fast (subject of this page) and Flash (the actual subject slug) must

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

Baseten's Aug 28, 2026 announcement for GLM 5.3 (the flagship, not the Flash subject) documents that the model is available through both Baseten Model APIs and Loops, has a 1M-token context window, defaults to high reasoning, and exposes a reasoning_effort parameter accepting low, high, or max to control reasoning dept This post is essential disambiguation context for the subject: it confirms the flagship GLM 5.3 is a separate SKU from GLM 5.3 Flash, so any feed item about the subject slug baseten-zai-orgglm-53-flash must not inherit flagship-level details. Loops-based fine-tuning is also surfaced as an option for the flagship, which

DataCamp provides an independent technical write-up characterizing GLM-5.3-Flash as Z.ai's cost-optimized sibling to the GLM-5.3 flagship, using a 320B-A18B mixture-of-experts configuration (320B aggregate parameters, 18B active per token), supporting a 1M-token context window, and being natively multimodal. The model The article also surfaces important deployment caveats relevant to anyone considering self-hosting: the FP8 checkpoint is approximately 306 GiB, ruling out lightweight local hosting, and the model ships at roughly 49 tokens/sec output speed. DataCamp notes Qwen3.8-Flash-Next launched the same day, framing Flash within

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

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

OpenRouter's listing for z-ai/glm-5.3-flash independently confirms an Aug 26, 2026 release, 1M context window, native multimodal support, and describes the model as suited for efficient coding and long-horizon agent tasks using a hybrid sparse and linear attention architecture. A limited-time 50% discount via ZAI is ac On Baseten specifically, OpenRouter measures a P50 latency of 1.09 seconds and throughput of 81 tps, with 99.56% uptime — figures that differ from Artificial Analysis's benchmark results because of differing workloads and methodologies, giving a second independent data point on Baseten's Flash performance alongside 19

Baseten announced on Aug 26, 2026 that GLM 5.3 Flash is available through Baseten Model APIs under the model identifier zai-org/GLM-5.3-Flash, reachable via Baseten's OpenAI-compatible endpoint with a Baseten API key. The supplied excerpt confirms GLM 5.3 Flash is Z.ai's latest model, features a 1M-token context window This changelog post is the primary source directly matching the subject model slug baseten-zai-orgglm-53-flash. The excerpt points developers to Baseten's documentation for Model APIs and to the dedicated GLM 5.3 Flash deployment for higher-throughput use cases, confirming both serverless and dedicated inference paths