GLM-5.3-Flash

GLM-5.3-Flash was confirmed by Bloomberg on August 26, 2026, as the identity of the anonymous "Ox Alpha" model that had been serving traffic on OpenRouter and OpenCode since August 20. The model comes from Zhipu AI (operating internationally as Z.ai), with 320 billion total parameters and 18 billion active per token. I During its six-day anonymous run on OpenRouter, Ox Alpha became the most-used model on the platform, processing approximately 23 trillion tokens — the biggest launch in OpenRouter's history at roughly 2.3 times the volume of the next model. On OpenCode it ended DeepSeek's 56-day run at the top of usage charts. Zhipu's

Z.ai released GLM-5.3-Flash on August 26, 2026, as the newest member of the GLM-5 family. The model has 320 billion total parameters with only about 18 billion active per token, supports multimodal inputs (text, image, video), offers a context window of up to 1 million tokens, and is available as an open-weight model u According to the article, GLM-5.3-Flash is the first natively multimodal model in the GLM-5 series and is designed around delivering frontier-level capabilities without frontier-level inference costs. Z.ai (formerly Zhipu AI) positions the model as a cost-efficient alternative for workloads where lower compute matters

Z.ai released GLM-5.3-Flash as the cost-optimized sibling to its flagship GLM-5.3, having arrived first under the anonymous preview codename "Ox Alpha." The model uses a 320B-A18B MoE design with a 1M token context window and is natively multimodal, scoring 84.3 on Terminal-Bench 2.1 — within striking distance of Claud The blended price for GLM-5.3-Flash is roughly $0.10 per 1M tokens, about a tenth of GLM-5.3's $0.90, with Artificial Analysis rating it at an intelligence index of 57 compared to 60 for the flagship GLM-5.3. Trade-offs include 49 tokens/sec output speed and a 306 GiB FP8 checkpoint that rules out lightweight local hos

On August 26, 2026, Z.ai released GLM-5.3-Flash as a 320-billion-parameter Mixture-of-Experts model that activates just 18 billion parameters per token, and it is the first natively multimodal model in the GLM-5 series. The release inverted the pattern set twelve days earlier when GLM-5.3 shipped on August 14 with its Architecturally, GLM-5.3-Flash holds a comparable total parameter count to the GLM-4.5 series (320B versus 355B) while nearly halving both activated parameters (18B versus 32B) and layer count (45 versus 92), driven by a hybrid attention stack that interleaves linear-attention layers with sparse-attention layers and ad

A direct comparison shows GLM 5.3 Flash differs from GLM 5.3 in several key dimensions. Flash is a 320B-total/18B-active MoE versus GLM 5.3's 753B-total/40B-active architecture, and Flash is the only one of the two that accepts native image input. Both share a 1-million-token context window and reasoning/tool-use capab GLM 5.3 Flash is priced at $0.15 per million input tokens and $0.50 per million output tokens, with output speed of approximately 49 tokens per second and time-to-first-token of 1.46 seconds. GLM 5.3 Flash is currently the only model in the comparison with publicly available open weights, offering 3× usable quota on Z.

The LLM Stats aggregator ranks GLM-5.3-Flash 17th on its composite score, placing it in the "Good / Top 10%" tier for Tool Calling (7 of 193) and "Average / Top half" for Reasoning (16 of 362), Chat (21 of 117), Vision (24 of 208), Coding (32 of 266), Math (38 of 327), Legal (76 of 210), and Finance (91 of 227). The bl Tracked benchmark scores include a GDPval-AA Elo of 1773 out of 3000 (rank 2), CharXiv-R of 0.89 (rank 8), and Terminal-Bench 2.1 results sourced from Z.ai and Artificial Analysis. A Quality Tracker shows performance shifting +0.44 sigma over the seven-day window versus a 30-day baseline of +6.71 sigma, with conversati

Z.ai shipped GLM-5.3-Flash on August 26, 2026, as a 320-billion-parameter mixture-of-experts model that activates only 18 billion parameters per token. It is the first natively multimodal model in the GLM-5 series, supporting text, image, and video inputs with a 1,048,576-token context window. The model runs natively i For the twelve days before launch, the model served traffic anonymously on OpenRouter and OpenCode under the name "Ox Alpha," becoming the most-used model on OpenRouter after processing roughly 23 trillion tokens. On benchmarks, GLM-5.3-Flash beats GLM-5.2 across six coding and agentic benchmarks at approximately one-t

GLM 5.3 Flash was released by Z.ai on August 26, 2026, with MIT-licensed weights on Hugging Face from day one. It is a 320-billion-parameter MoE that activates 18 billion parameters per token, making it the first natively multimodal model in the GLM-5 series and the first built on a new base since GLM 5.2. It supports The model is priced at $0.15 per million input tokens and $0.50 per million output tokens, compared with $1.40 and $4.40 for the larger GLM 5.3. In independent Artificial Analysis testing, GLM 5.3 Flash scores 57 on the Intelligence Index compared with 60 for GLM 5.3, while activating only 18B parameters versus 40B for