@KyleHessling1: Made an RTS game with our soon-to-release Qwopus-Coder-35B-A3B entirely in @opencode Thinking enabled, thinking cap set…
Summary
Kyle Hessling announces the upcoming release of the Qwopus-Coder-35B-A3B coding model, demonstrating its capability by using it with OpenCode to develop a fully functional real-time strategy game. The model achieves high speed and draft acceptance on a GeForce RTX 5090.
View Cached Full Text
Cached at: 06/27/26, 05:53 AM
Made an RTS game with our soon-to-release Qwopus-Coder-35B-A3B entirely in @opencode
Thinking enabled, thinking cap set at 2048 tokens to prevent runaway, –temp set to 1.
Getting crazy speeds in single stream with a 5090: Averaged 96.7% MTP draft acceptance for 253tps!
I have put together a full server recommendations file, along with the ideal opencode configuration I used for this run, it will all be present in the card on launch of the model! In the meantime, you can play the game in my HF space, linked in comments!
We’re finalizing some model recommendations and will get the model card built for release in the next few days.
This game was a far more ambitious prompt than usual for demos, and I genuinely didn’t think it would work, but we have a fully playable RTS with a decent featureset, could keep iterating and building on top of it, or anyone is welcome to take the source code from HF when it’s live and use it as a template for your own ideas!
It’s surprisingly fun and difficult. I am very impressed that the enemy AI is as competent as it is. I lost in this quick test, but you can see how functional everything is.
Jinja template fixes from Qwopus-Compat will be included by default!
Looking forward to seeing what you all build with the speed and capability of this new model; we will try to have it live as soon as possible!
Happy Friday, everyone! Thanks for reading this and messing with our models! Grateful for all of the feedback and support!
Similar Articles
@KyleHessling1: Good morning y'all! Qwopus-3.6-35B-A3B-MTP-Coder is live! All GGUF's will be populating over the next few hours! It's a…
Qwopus-3.6-35B-A3B-MTP-Coder is a new open-source MoE fine-tune optimized for coding agent workflows with thinking disabled, offering fast token-efficient inference and competitive performance against similar models.
@loktar00: Not sure how Kyle doesn't have 10's of thousands of followers. He's putting out incredibly cool usable stuff what seems…
Kyle Hessling announces the release of Qwopus-3.6-35B-A3B-MTP-Coder, a fast MOE model specialized for coding, available in GGUF format.
@sudoingX: update: qwen 3.6 27b dense q4 just one shotted octopus invaders game on a single 3090. hermes agent drove the whole thi…
A user benchmark demonstrates that the Qwen 3.6 27B dense model (Q4 quantized) can autonomously generate a fully playable multi-file game in a single prompt on a single RTX 3090, significantly outperforming its predecessor with zero manual interventions. The results highlight major improvements in local code generation and agentic capabilities for consumer-grade hardware.
@KyleHessling1: Hello again, everyone! We've got another really fun 9b, this one specifically trained for tool calling and agentic codi…
A new 9B fine-tuned model called Qwopus3.5-9B-Coder is released, optimized for tool calling and agentic coding workflows, achieving strong SWE-bench and HermesAgent-20 scores while running on affordable hardware.
@outsource_: BREAKING QWOPUS 3.6 27B IS FULLY LIVE! SOTA QWEN 3.6 27b + Opus IS HERE!!!! Agentic coding GOATED: 75.25% (152/202) on …
Qwopus 3.6 27B is now fully live, a merged model (Qwen + Opus) achieving state-of-the-art agentic coding performance with 75.25% on SWE MMLU Pro, handling 303k token context at Q8 KV cache, and running on 24GB VRAM at Q5_K_M quantization.