Opencode killed the $5 first-month promo on its Go plan. First month is $10 flat now. Their cost pressure must be real — they also just dropped the free DS-V4-Flash-Free model.
Opencode killed the $5 first-month promo on its Go plan
More posts
meta/muse-spark-1.2-contributor is free to use.opencode's free
tier also has meta/muse-spark-1.2-contributor, but that model's server is flaky, so I use hy-3 as my subagent model most of the time. Also, the opencode free models on 9router don't auto-show in pi. You can only nest the oc free model via the composite form, and that combo caps at 128K context unless you hand-write the config. So I built a pi extension: set /9router with your address and key, and it auto-detects every model on 9router, including the oc free ones. pi install npm:@haoyiyin/9router
muse-spark-1.2-contributor seems really bad at following Chinese
prompts? Compared notes with a few people, same impression across the board. It keeps misreading what you ask in Chinese. Some say Meta is quietly nerfing the version it serves Chinese users. Maybe. I think the model's just weak in Chinese.
Meituan shipped LongCat-2.5
. 1.6T params, ~48B active, native multimodal, aimed at long-horizon agents, trained start to finish on Chinese hardware. Images get quantized into multi-level discrete tokens, and that's where Visual Coding and 3D pick up. On the RL side, failed trajectories get turned back into verifiable training tasks. W4A4 inference, 43% faster prefill. Meituan's own evals put it between Claude Opus 4.8 and Opus 5.
SenseTime open-sourced SenseNova Loop-9B-Preview.
It stacks layers instead of parameters. The decoder has 32 layers, but layers 4 through 27 run twice with shared weights, so a forward pass does 56 layers of work while the checkpoint stays at 9B. Their own eval numbers: 71.38 on BIRD-Dev and 13.72 on AutomationBench, both above Qwen3.5-27B. Terminal Bench is still behind.
