GLM 5.3 Flash

Yotta Labs' comparison piece documents that Z.ai shipped GLM 5.3 on August 14, 2026 as an API-only flagship, followed by GLM 5.3 Flash on August 26 with MIT-licensed weights on Hugging Face, text/image/video input, and launch pricing around one-tenth the flagship's. The piece highlights a key methodological issue: the On that apples-to-apples comparison, Flash lands within 3.5 points of the flagship at a tenth of the price, with open weights that the flagship lacks. The piece notes that the flagship's 744B weights remain unreleased, while Flash spent a week running in stealth as "Ox Alpha" before Z.ai claimed it, launching with real

The Medium summary (August 26, 2026) by Greek Ai on CodeToDeploy restates the core GLM-5.3-Flash launch facts: 320 billion total parameters with about 18 billion active, multimodal input support, a 1M-token context window, and open weights under the MIT license, all released by Z.ai on August 26, 2026. The framing posi The piece is largely a synopsis of Z.ai's announcement text rather than an independent technical analysis. It repeats the value proposition of dramatically lower inference costs while delivering competitive performance, but adds limited new benchmark or architectural detail beyond what the launch announcement itself pr

Ampere's head-to-head (August 27, 2026) cites Artificial Analysis intelligence-index scores of 60 for GLM 5.3 and 57 for GLM 5.3 Flash, alongside output-speed measurements of roughly 85 tokens per second for the flagship versus about 49 tok/s for Flash. The comparison frames Flash as dramatically cheaper and natively m The spec table lists GLM 5.3 as a 753B-total / 40B-active MoE and Flash as 320B-total / 18B-active, both with a 1M context window. Pricing is $1.40/M input and $4.40/M output for the flagship against $0.15/M input and $0.50/M output for Flash, with Flash weights available now and the flagship's weights listed as "comin

The AI Release Tracker confirms GLM-5.3-Flash as an open-weight model released by Z.ai on August 26, 2026, twelve days after GLM-5.3, with a 1M token context window and 320B parameters. Its benchmark coverage spans Toolathlon-Verified, CharXiv Reasoning, OfficeQA Pro, MVBench, MMVU, DeepSWE 1.1, and nine additional ben Provider pricing tiers reveal significant spread, from DeepInfra and GMICloud at $0.075/$0.25 per million tokens (fp4 and fp8 respectively) through mid-range options at $0.10–$0.14 per million input, to the Z.ai canonical rate of $0.15/$0.50. Context windows range from 256K (Io Net, Reka) to 1M for most providers, with

Z.ai released GLM 5.3 Flash on August 26, 2026 as a 320-billion-parameter mixture of experts that activates just 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. The model was revealed to be "Ox Alpha," the anonymous list MIT-licensed weights landed on Hugging Face on day one, with API pricing set at $0.15 per million input tokens and $0.50 per million output tokens, compared to $1.40 and $4.40 for the full GLM 5.3 flagship. The analysis notes that Z.ai's promised open-weight release of GLM 5.3's 744B parameters remains outstanding, wit

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, 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, accepting text, image, and video inpu Weights were released on Hugging Face under an MIT license on day one, with the API priced at $0.15 per million input tokens and $0.50 per million output tokens, and a public vLLM recipe for self-hosting. According to the page's benchmark sheet, GLM-5.3-Flash beats GLM-5.2 across six coding and agentic benchmarks at ro

LLM Stats ranks GLM-5.3-Flash 18th on its composite LLM Stats Score, placing it in the "A Good" tier (top 10%) on Chat (3 of 117) and Tool Calling (7 of 194), while remaining in the "B Average" tier on Reasoning (17 of 363), Vision (24 of 208), Coding (32 of 267), Math (38 of 327), Legal (70 of 211), and Finance (86 of The page's quality tracker records a +0.89σ improvement over seven days, +0.79σ over thirty days, and +8.43σ versus baseline, tracking the model's performance trajectory from late August through mid-September 2026. Benchmark scores include GDPval-AA Elo 1773/3000 (rank 2), CharXiv-R 0.89/1 (rank 8), and Terminal-Bench

BenchLM's tracking page for GLM-5.3-Flash places the model at an overall capability score of 66.1/100 (field median 56.3), ranking it 32nd among 230 tracked models. It records particularly strong multimodal and grounded performance, ranking 10th in that category, with validated strength for screenshots, documents, char The page records the model's 1M-token context window (field median 256,000) and notes that output speed and time-to-first-token are not measured for this entry, with 19 published benchmark rows leaving some tracked slots empty. No comparable first-party hosted token rate is published on this page. Each displayed score