dropkit.contents

Ledger on GitHub. Hub here. Branches elsewhere.

← Drops · 2026-09-14

drop-2026-09-14-us-am

claimlabelordernumbersURL
Qwen3.8-Max shipped as a sparse MoE flagship with 2.4T total parameters and 95B active per token, with a 1M-token context window.🟢 robust12.4T; 95B; 1Mhttps://www.digitalapplied.com/blog/qwen3-8-max-full-release-benchmarks-open-weights
Alibaba later published open weights for Qwen3.8-2.4T-A95B, described as the open-weight version of Qwen3.8-Max.🟢 robust22.4T; 95Bhttps://aws.amazon.com/blogs/machine-learning/deploying-qwen3-8-2-4t-a95b-on-amazon-sagemaker-hyperpod-with-vllm/
The open Qwen3.8-2.4T-A95B checkpoint was published on Hugging Face with 262,144 native tokens and extensibility up to 1,010,000 tokens.🟢 robust3262,144; 1,010,000https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B
DeepSeek V4 Pro was priced at $1.32 per million input tokens and $3.96 per million output tokens, with peak/off-peak pricing introduced in mid-August 2026.🟢 robust4$1.32/M input; $3.96/M outputhttps://www.reuters.com/world/china/deepseek-releases-official-v4-pro-model-it-steps-up-expansion-2026-08-13/
DeepSeek V4 Pro was also described in secondary coverage as a sharp inference-cost squeeze, with a prior flat output rate of $0.87/M used as the comparison baseline.⚠️ sensitive5$0.87/Mhttps://www.techtimes.com/articles/324764/20260817/deepseek-v4-api-prices-quadruple-peak-what-developers-pay-starting-now.htm
GLM-5.3-Flash was described as a 321B-total / 18B-active multimodal model with 1M-token context and MIT licensing.🟢 robust6321B; 18B; 1Mhttps://ollama.com/library/glm-5.3-flash
GLM-5.3-Flash allegedly scored 57/100 on the Artificial Analysis Intelligence Index, placing it in frontier territory.⚠️ sensitive757/100https://www.techtimes.com/articles/325872/20260828/sanctioned-chinese-chips-just-served-62-trillion-ai-tokens-frontier-scale.htm
ByteDance was said to be pre-training a 10T-parameter model.⚠️ sensitive810Thttps://www.ainchina.com/blog/china-ai-model-wars-summer-2026/
China’s August 2026 model wave was summarized as five frontier models in eight weeks.⚠️ sensitive95 models; 8 weekshttps://www.ainchina.com/blog/china-ai-death-zone-five-models-eight-weeks-2026/
August 2026 was described as the densest month of model releases, with trackers citing 14–24 confirmed releases.⚠️ sensitive1014–24 releaseshttps://www.buildfastwithai.com/blogs/ai-news-today-august-30-2026
DeepSeek, MiniMax, Alibaba, and Zhipu were all said to have joined the open-weight release wave through August 2026.🟢 robust114 labs namedhttps://www.chosun.com/english/opinion-en/2026/08/24/XHX6TUKL3RBMFMOZL5QW5NQW2U/