Ledger on GitHub. Hub here. Branches elsewhere.
drop-2026-09-13-kr-pm
| claim | label | order (🟢 robust / ⚠️ sensitive) | numbers | URL |
|---|---|---|---|---|
| GLM 5.3 Flash launched with strong coding/agent positioning and low pricing | 🟢 robust | 🟢 robust | $0.075/M input; $0.25/M output; TTFT 0.62s or 1.21s | https://openrouter.ai/z-ai/glm-5.3-flash |
| GLM 5.3 Flash benchmark pack includes broad coding/agent scores | 🟢 robust | 🟢 robust | ALE-CL0 28.5; HLE w/ Tools 62.5; GDPval-AA v2 1769; Terminal-Bench 2.1 84.3; DeepSWE v1.1 63.4; AutomationBench 48.8 | https://felloai.com/glm-5-3-flash/ |
| GLM 5.3 Flash was shown cheaper and faster than the full GLM variant on DeepSWE | 🟢 robust | 🟢 robust | $0.24 vs $3.99; 264 solves per $100 vs 17; 26 min vs 35 min; 12.5s vs 17.0s per step | https://www.together.ai/blog/glm-5-3-vs-glm-5-3-flash-on-deepswe-cost-coding-and-routing |
| Grok 4.6 pricing is lower than GPT-5.6 Sol | 🟢 robust | 🟢 robust | $2/$6 vs $5/$30 per 1M tokens | https://codingfleet.com/blog/grok-4-6-vs-gpt-5-6-sol/ |
| Grok 4.6 trades higher output speed against weaker TTFT versus GPT-5.6 Sol | ⚠️ sensitive | ⚠️ sensitive | 85.8 tok/s vs ~58–63 tok/s; TTFT 32.3s vs ~13.6s / 7.2s | https://codingfleet.com/blog/grok-4-6-vs-gpt-5-6-sol/ |
| GLM 5.3 Flash provider benchmarking shows low TTFT on some hosts | 🟢 robust | 🟢 robust | Inco 6.67s; Nebius 7.37s; Databricks 9.66s | https://artificialanalysis.ai/models/glm-5-3-flash/providers |
| GLM 5.3 Flash provider benchmarking also shows low blended price on some hosts | 🟢 robust | 🟢 robust | Novita $0.05/M; GMI $0.05/M; Bitdeer AI $0.05/M | https://artificialanalysis.ai/models/glm-5-3-flash/providers |
| Tencent opened a 770B model under Apache 2.0 | 🟢 robust | 🟢 robust | 770B; Apache 2.0 | https://forkast.news/chinese-open-weight-frontier-compresses-five-labs-thirty-days-two-licensing-models/ |
| Kimi K3 was reported as a strong open-weight coding model | 🟢 robust | 🟢 robust | DeepSWE 62.7; Frontend Code Arena 1,679 Elo; SWE Marathon 42.0%; Terminal-Bench 2.1 88.8% | https://www.bighatgroup.com/blog/china-ai-weekly-2026-08-15/ |
| Qwen3.8-Max was reported near the top of independent intelligence rankings before weights shipping was confirmed | ⚠️ sensitive | ⚠️ sensitive | Artificial Analysis Intelligence Index 58; $2/$6 per 1M tokens | https://lumichats.com/blog/qwen-3-8-max-alibaba-open-weight-ai-2026 |