GLM-5.3-Flash from Z.ai (the Beijing lab also known as Zhipu AI) was revealed on August 26, 2026 as the model previously served anonymously on OpenRouter and OpenCode under the name "Ox Alpha." According to this account, Zhipu confirmed via Bloomberg that Ox Alpha was GLM-5.3-Flash, a 320-billion-parameter mixture-of-e The piece frames GLM-5.3-Flash as a frontier-capability open-weights model running without NVIDIA chips, an explicit positioning that ties the release to U.S.-China compute competition and developer accessibility. It highlights OpenRouter topping usage charts and notes Stripe's agreement to acquire OpenRouter and CEO P

This technical breakdown details Z.ai's August 26, 2026 release of GLM 5.3 Flash as a 320B-total, 18B-active Mixture-of-Experts model — not a distilled version of GLM 5.3 but a new model trained from a newly designed base, with hybrid sparse-plus-linear attention. It is the first natively multimodal entry in the GLM-5 The post confirms the Ox Alpha identity reveal: GLM 5.3 Flash is the anonymous model that dominated OpenRouter and OpenCode usage for roughly twelve days before publication. It walks through a full benchmark table including the tests Z.ai's chart omits, provides a brief inference-cost comparison versus GLM 5.2, and cle

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

Independent Artificial Analysis testing rates GLM 5.3 at 60 on the Intelligence Index versus 57 for GLM 5.3 Flash, but GLM 5.3 Flash costs roughly one-ninth as much at normal API pricing and activates only 18B parameters versus 40B for the flagship. Surprisingly, GLM 5.3 Flash is not faster at token generation: Artific The comparison highlights that GLM 5.3 Flash is the better value for most high-volume applications and multimodal agents, with native image input, visual coding, browser-agent, and computer-use-agent wins in head-to-head testing. Specifications confirm both models come from Z.ai, share a 1M token context window, suppor

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 18B per token, runs natively in FP8, and supports a 1,048,576-token context window. It is the first natively multimodal model in the GLM-5 series. For twelve days before launch it had been running anon The deep-dive emphasizes that GLM-5.3-Flash is not a trimmed version of the flagship but a separate model trained from a new base checkpoint, with its architecture and training recipe redesigned for efficiency. It beats GLM-5.2 across six coding and agentic benchmarks at roughly one-tenth of the price and approaches Cl