Ledger on GitHub. Hub here. Branches elsewhere.
drop-2026-09-07-us-am
| claim | label | order | numbers | URL |
|---|---|---|---|---|
| DeepSeek V4-Pro was generally available on app, web, and API by Aug 13, 2026. | 🟢 confirmed | 🟢 robust | Aug 13, 2026 | https://api-docs.deepseek.com/news/news260813/ |
| DeepSeek V4-Pro pricing increased versus prior V4 Flash pricing, with Reuters citing $1.32/M input and $3.96/M output. | 🟢 confirmed | 🟢 robust | $1.32/M input; $3.96/M output; ~9x / ~14x vs V4 Flash | https://www.reuters.com/world/china/deepseek-releases-official-v4-pro-model-it-steps-up-expansion-2026-08-13/ |
| DeepSeek V4-Pro kept a premium flagship position tied to stronger benchmark performance. | 🟡 partial | ⚠️ sensitive | premium positioning; benchmark-driven pricing | https://www.reuters.com/technology/deepseek/ |
| DeepSeek-V4-Pro was framed as improving agent capabilities and keeping a 1M-token context window. | 🟡 partial | ⚠️ sensitive | 1M context | https://www.unite.ai/deepseek-ships-v4-pro-as-its-flagship-model-leaves-preview/ |
| Qwen3.8-Max was officially launched on Aug 3, 2026. | 🟢 confirmed | 🟢 robust | Aug 3, 2026 | https://qwen.ai/research |
| Qwen3.8-Max was described as a 2.4T-parameter MoE with 95B active parameters. | 🟢 confirmed | 🟢 robust | 2.4T total; 95B active | https://www.marktechpost.com/2026/08/03/alibaba-qwen-releases-qwen3-8-max/ |
| Qwen3.8-Max was positioned as Alibaba’s largest and most capable Qwen model to date. | 🔴 hype | ⚠️ sensitive | n/a | https://qwen.ai/research |
| Qwen3.8-Max was reported with 262,144 native context and benchmark scores like GPQA Diamond 92.6 and SWE-bench Pro 67.7. | 🟡 partial | ⚠️ sensitive | 262,144 context; 92.6; 67.7; 86.6; 93.0 | https://www.artifocial.com/blog/ai-news-2026-aug-10-16 |
| GLM-5.3 was released on Aug 14, 2026 by Z.ai/Zhipu. | 🟢 confirmed | 🟢 robust | Aug 14, 2026 | https://www.unite.ai/z-ai-launches-glm-5-3-with-frontier-coding-and-a-cyber-capability-that-outgrew-its-training/ |
| GLM-5.3 was reported to keep the same base model as GLM-5.2 and derive gains from post-training. | 🟡 partial | ⚠️ sensitive | same base model; post-training gains | https://www.unite.ai/z-ai-launches-glm-5-3-with-frontier-coding-and-a-cyber-capability-that-outgrew-its-training/ |
| GLM-5.3 weights were not yet public at release and were said to be held back for safety evaluation. | 🟡 partial | ⚠️ sensitive | weights delayed ~2 weeks | https://www.unite.ai/z-ai-launches-glm-5-3-with-frontier-coding-and-a-cyber-capability-that-outgrew-its-training/ |
| GLM-5.3 family details were reported as roughly 743B–753B total MoE parameters, about 40B active, with 1M-token context. | 🟡 partial | ⚠️ sensitive | 743B–753B total; ~40B active; 1M context | https://www.apidog.com/blog/self-host-glm-5-3-open-weights/ |
| GLM-5.3-Flash was later described as a 320B total / 18B active variant with weights on Hugging Face. | 🟡 partial | ⚠️ sensitive | 320B total; 18B active | https://aitoolsrecap.com/Blog/glm-53-open-weights-release-2026 |
| DeepSeek Harness was released as an open-source agent harness aimed at Claude Code-style workflows. | 🟡 partial | ⚠️ sensitive | n/a | https://buttondown.com/JacopoCastellano/archive/ai-digest-2026-08-15/ |
| Qwen ecosystem reportedly surpassed 3B downloads globally, with about 2.045B on Hugging Face alone. | 🟡 partial | ⚠️ sensitive | 3B global; 2.045B HF | https://www.requesty.ai/blog/open-weight-frontier-august-2026-glm-qwen-hy4 |
| OpenAI |