OpenCode Go Updates (18/11/2026)
programming.dev
OpenCode Go Updates (18/11/2026)
Last three weeks included the following changes: ## Go rate limits tweaked OpenCode Go revised their rate limits, they no longer provide explicit weekly or monthly rate limits, only the 5 hours one. And instead of the detailed list before now they say in their docs [https://opencode.ai/docs/console/go/#usage-limits]: > The 5-hour limit is 20% of your monthly allowance. The window starts when you first use your allowance and lasts five hours. > The weekly limit is 50% of your monthly allowance. It resets on Monday at 00:00 UTC. ## Updates to Go plan * (Boosted) GLM-5.3 usage raised from $15 to $100! * (Boosted) DeepSeek V4.1 Flash usage raised from $80 to $100 * (New) Qwen4-Max the big daddy of Qwen3.8-FN. This one costs half compared to Qwen3.8 Max so now it’s actually somewhat usable for Orchestration at least * (New) Hy4 the full release of Hy4 (it was in preview for a while) Model |5HOURS |1WEEK |1MONTH |CREDITS -----------------------------|-------|-------|---------|--------- GLM-5.3 (++usage) | 1320 | 3300 | 6600 | 100 DeepSeek V4.1 Flash (++usage) | 43330 | 108325 | 216650 | 100 Qwen4 Max |463 |1158 | 2315 | 15 Hy4 | 2710 | 6775 | 13550 | 60 While aside, here’s some important happening: ## Artificial Analysis fails Artifical Analysis [https://artificialanalysis.ai/] “tweaked” their benchmarks and scoring to push all Chinese LLMs down and create a virtual moat between them and the “sanctioned” models. Interesting timing to do this. They ended up in a situation where certain big-tech models are ahead by 10+ points compared to the top Chinese LLMs that outperform them in actual professional usage. Suffice it to say, it’s no longer an accurate score to evaluate models and I’m sad to see benchmarking become a political tool, although with the bench maxxing situation the benchmarks weren’t that useful anyway. ## Search for Orchestrator I’ve personally been looking for a good Orchestrator in my current pipeline, GLM-5.3 is favorite for this but $15 was too little and ran out very quickly taking down the entire Go plan with it so I started using GPT-6 Astra which also ran into OpenAI’s daily rate limits even as an orchestrator (low in/out tokens). Then switched to GPT-6.1 Sol which is horrible as an orchestrator for some reason and actually indirectly caused issues in the code base that I’m still dealing with. I was gonna try GPT-6.1 Terra next but now GLM-5.3 on Go with its $100 quota is the new king of orchestrators. Another thing that happened over the last two days (although not opencode related) is that the Qwen token plans got a significant boost to usage quotas. Enough that I was considering moving to their $10 plan to use Qwen4-Max as Orchestrator. However now that GLM-5.3 is definitely usable in Go I’m probably staying with them. Good competition 💪
0 comments
No comments yet.