Ledger on GitHub. Hub here. Branches elsewhere.
drop-2026-09-08-kr-close
| claim | label | order (🟢 robust / ⚠️ sensitive) | numbers | URL |
|---|---|---|---|---|
| Grok 4.6 API list price is $2 input / $6 output per 1M tokens, with a fast variant at 2× price and a 200K+ long-context band. | 🟢 confirmed | 🟢 robust | $2 / $6; $4 / $12; 200K | https://x.ai/news/grok-4-6 |
| Grok 4.6 benchmarked at AA Intelligence Index 61, tying GPT-5.6 Sol, while Opus 5 peaked at 63. | 🟢 confirmed | 🟢 robust | 61; 63 | https://x.com/ArtificialAnlys/status/2087564648325530099 |
| Grok 4.6 appeared in Cursor and Grok Build on day one. | 🟡 partial | ⚠️ sensitive | day-one | https://byteiota.com/grok-4-6-is-here-what-developers-actually-get/ |
| GPT-5.6 Sol API was priced at $5 input / $30 output per 1M tokens in the same AA comparison. | 🟢 confirmed | 🟢 robust | $5 / $30 | https://x.com/ArtificialAnlys/status/2087564648325530099 |
| Claude Opus 5 coding/agent pricing was listed at $5 input / $25 output per 1M tokens. | 🟢 confirmed | 🟢 robust | $5 / $25 | https://www.stackspend.app/blog/model-selection/best-coding-models-august-2026 |
| Anthropic Computer Use, Browser Use, Skills API, and Files API were described as GA in a third-party roundup, but primary docs still marked computer use beta. | 🔴 hype | ⚠️ sensitive | 4 tools | https://console.anthropic.com/llms-full.txt |
| Anthropic Flash-tier pricing was reported as $0.75 input / $3.75 output per 1M tokens. | 🟡 partial | ⚠️ sensitive | $0.75 / $3.75 | https://www.unrot.co/blogs/today-top-ai-news-august-25-2026 |
| MCP was reported as updated on 2026-07-28 with stateless-server language in a weak secondary source. | 🟡 partial | ⚠️ sensitive | 2026-07-28 | https://www.unrot.co/blogs/today-top-ai-news-august-25-2026 |
| Qwen3.8-Max launched as Alibaba’s flagship with 2.4T total parameters, 95B active parameters, and 1M-token context. | 🟢 confirmed | 🟢 robust | 2.4T; 95B; 1M | https://www.alibabacloud.com/en/press-room/alibaba-unveils-qwen3-8-max?_p_lc=1 |
| Qwen3.8-Max hosted API pricing was reported at $2 input / $6 output per 1M tokens. | 🟡 partial | ⚠️ sensitive | $2 / $6 | https://www.marktechpost.com/2026/08/03/alibaba-qwen-releases-qwen3-8-max/ |
| Qwen3.8-Max terminal, paper, and SWE results were reported as 86.6, 93.0, and 67.7 respectively, though independent comparison differed. | 🟡 partial | ⚠️ sensitive | 86.6; 93.0; 67.7 | https://www.marktechpost.com/2026/08/03/alibaba-qwen-releases-qwen3-8-max/ |
| Tencent Hy4 preview was open-sourced with 770B total parameters, 49B active parameters, and 1M+ context. | 🟢 confirmed | 🟢 robust | 770B; 49B; 1M+ | https://www.tencent.com/tencent-releases-and-open-sources-tencent-hy4-preview/ |
| Hy4 hosted pricing was reported at about $0.83 per 1M input tokens. | 🟢 confirmed | 🟢 robust | $0.834 in; $2.501 out; $0.042 cache | https://hy.tencent.ai/research/hy4-preview |
| DeepSeek V4 Pro pricing was reported as $1.32 input / $3.96 output per 1M tokens at peak, with off-peak discounts. | 🟡 partial | ⚠️ sensitive | $1.32 / $3.96; $0.66 / $1.98 | https://www.reuters.com/world/china/deepseek-releases-official-v4-pro-model-it-steps-up-expansion-2026-08-13/ |
| China’s average inference price during Aug. 6–8 was reported at about US$1.16–US$1.18 per million tokens. | 🟡 partial | ⚠️ sensitive | $1.16– |