Ledger on GitHub. Hub here. Branches elsewhere.
drop-2026-09-10-kr-am
| claim | label | order (🟢 robust / ⚠️ sensitive) | numbers | URL |
|---|---|---|---|---|
| Alibaba’s Qwen3.8-Max is a 2.4T-parameter open-weight model with about 95B active parameters per token, and Reuters reports API pricing at $2 per million input tokens and $6 per million output tokens. | 🟢 confirmed | 🟢 robust | 2.4T total; ~95B active; $2 / $6 per 1M tokens | https://www.reuters.com/business/retail-consumer/alibabas-qwen-launches-qwen38-flash-ai-model-with-lower-training-costs-2026-08-26/ |
| The Qwen3.8 open-weight release is described as a sparse MoE system with 512 experts, 10 routed plus 1 shared active per token, and a native context window of 262,144 tokens expandable to 1,010,000 tokens. | 🟢 confirmed | 🟢 robust | 512 experts; 10+1 active; 262,144 → 1,010,000 ctx | https://www.reuters.com/business/retail-consumer/alibabas-qwen-launches-qwen38-flash-ai-model-with-lower-training-costs-2026-08-26/ |
| DeepSeek formally released V4 Pro on Aug. 13 and Reuters says its API prices increased by 50% to 1,100% depending on model, token type, and time of use. | 🟢 confirmed | 🟢 robust | 50%–1,100% increase | https://www.reuters.com/world/china/deepseek-raises-api-pricing-its-v4-models-2026-08-13/ |
| Reuters reported DeepSeek V4-Pro-0813 pricing at $1.32 per million input tokens and $3.96 per million output tokens. | 🟢 confirmed | 🟢 robust | $1.32 / $3.96 per 1M tokens | https://www.reuters.com/world/china/deepseek-releases-official-v4-pro-model-it-steps-up-expansion-2026-08-13/ |
| Alibaba released Qwen3.8-Flash with a 125B-parameter model and Reuters says it is performance-competitive with recent rival releases. | 🟢 confirmed | 🟢 robust | 125B | https://www.investing.com/news/stock-market-news/alibaba-releases-qwen38flash-ai-model-to-cut-costs-93CH-4877431 |
| Reuters says Qwen3.8-Flash has a default 262,144-token context window expandable to 1 million tokens. | 🟢 confirmed | 🟢 robust | 262,144; 1,000,000 | https://www.reuters.com/business/retail-consumer/alibabas-qwen-launches-qwen38-flash-ai-model-with-lower-training-costs-2026-08-26/ |
| Reuters says Alibaba’s Qwen service can be connected by Apple users in China, showing distribution through Apple-related channels. | 🟡 partial | ⚠️ sensitive | none stated | https://www.reuters.com/business/retail-consumer/apple-says-mac-users-china-can-connect-alibabas-qwen-ai-service-2026-08-08/ |
| Reuters’ company coverage describes Alibaba’s Qwen3.8-Max as open-source and open-weight. | 🟢 confirmed | 🟢 robust | none stated | https://www.reuters.com/company/alibaba-group-holding-ltd/ |
| DeepSeek V4 Pro is framed by Reuters as a premium flagship move to compete with domestic rivals, but the snippet does not provide benchmark figures. | 🟡 partial | ⚠️ sensitive | none given | https://www.reuters.com/world/china/deepseek-raises-api-pricing-its-v4-models-2026-08-13/ |