Ledger on GitHub. Hub here. Branches elsewhere.
drop-2026-09-13-us-am
| claim | label | order | numbers | URL |
|---|---|---|---|---|
| GLM 5.3 Flash is a China-side open-weight MoE with multimodal input and a 1M-token context window | 🟢 robust | 🟢 robust | 320B total; 18B active; 1M context | https://venice.ai/models/z-ai-glm-5-3-flash |
| Qwen3.8-Max shipped as Alibaba’s largest model with long context and delayed open weights | 🟢 robust | 🟢 robust | 2.4T parameters; 95B active; 1M context | https://www.qwencloud.com/models/qwen3.8-2.4t-a95b |
| OpenAI’s Jalapeño is a custom inference chip aimed at serving-scale efficiency | 🟢 robust | 🟢 robust | 1.7 exaFLOPS; 27.5 TB HBM4; ~2 PB/s bandwidth | https://www.theregister.com/systems/2026/08/25/openais-upcoming-jalapeno-chip-looks-like-itll-be-an-inference-beast/5292052 |
| OpenAI said Jalapeño would begin deployment in its infrastructure by end of year | ⚠️ sensitive | ⚠️ sensitive | end of 2026 deployment; generations 2 and 3 in progress | https://www.cnbc.com/2026/08/26/openai-jalapeno-ai-chip-nvidia.html |
| Qwen3.8-Flash exists as a multimodal Qwen family release with a lower-price API | ⚠️ sensitive | ⚠️ sensitive | $0.16/M input; $0.47/M output | https://docs.qwencloud.com/changelog/models |
| Chinese open models were described as taking a larger share of global downloads than US open models | ⚠️ sensitive | ⚠️ sensitive | no exact figure verified here | https://www.aimodeling.com/en/news/slug/hugging-face-summer-2026-frontier-ceiling |
| OpenAI’s Jalapeño benchmarks were presented as better than Nvidia-class hardware on serving efficiency | ⚠️ sensitive | ⚠️ sensitive | up to 3.6× speed; up to 1.9× work-per-watt | https://www.axios.com/2026/08/25/openai-says-its-jalapeno-chip-offers-spicy-performance |
| Qwen3.8-Max was reported to have open weights arriving the following week after launch | ⚠️ sensitive | ⚠️ sensitive | open weights next week | https://www.alibabacloud.com/blog/alibaba-unveils-qwen3-8-max-its-largest-and-most-capable-flagship-model-to-date_603420 |
| GLM 5.3 Flash launch pricing was framed as far cheaper than flagship alternatives | ⚠️ sensitive | ⚠️ sensitive | $0.15/M input; $0.50/M output | https://www.felloai.com/glm-5-3-flash/ |
| OpenAI Jalapeño was described as a “threat” to Nvidia margins | ⚠️ sensitive | ⚠️ sensitive | no verified number beyond the benchmark claims | https://www.cnbc.com/2026/08/26/openai-jalapeno-ai-chip-nvidia.html |