Qwopus
launchlate March 2026
powered byQwopus 3.5 27B-v3, Qwopus 3.5 9B-v3
goblin vibe check:
good call if you want a local model that shows its work and doesn't hallucinate function calls as much
reasoning-distilled qwen-based model family tuned for clearer internal planning, stronger coding behavior, and more reliable tool use.
context
262k
tokens
Reasoning distilled from Claude Opus chainsLower think-token cost than earlier distilled variantsTool-calling reinforcement learning for agent workflowsStrong front-end and coding stability versus base Qwen builds
key features
Reasoning distilled from Claude Opus chainsLower think-token cost than earlier distilled variantsTool-calling reinforcement learning for agent workflowsStrong front-end and coding stability versus base Qwen builds
spec & usage
Post-trained Qwen 3.5 causal model using an act-then-refine workflow instead of excessive pre-action overthinking
Weights are Apache 2.0 and optimized for long autonomous local runs on single 3090/4090-class GPUs
Best suited for codebases, long documents, and tool-using agents that need native long context
limitations
Chain-of-thought is still longer than previous iterations
Can still hallucinate external facts when the context is thin
scope:
codelanguageagentresearchlocalopen-sourcefree
launchlate March 2026