Jackrong/Qwopus3.8-27B-Flash-V2

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 15, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Jackrong/Qwopus3.8-27B-Flash-V2 is a 27 billion parameter language model, post-trained from Qwopus3.8-27B-Flash, which is based on Qwen3.8-27B. This model is optimized for practical inference efficiency and agentic workflows, aiming to reduce ineffective reasoning and accelerate task completion. It features improved Python formatting and is designed for resource-constrained environments where wall-clock time and token cost are critical. The model excels in agentic software-engineering tasks and local MTP speculative decoding.

Loading preview...

Model Overview

Jackrong/Qwopus3.8-27B-Flash-V2 is a 27 billion parameter model, building upon the Qwen3.8-27B foundation and further post-trained from Qwopus3.8-27B-Flash. Its core philosophy is to preserve strong capabilities while collapsing inference costs, making it highly suitable for demanding agent workloads. The V2 iteration applies revised reward functions and reinforcement learning methods to enhance efficiency and consistency, aiming for quicker, cleaner task completion.

Key Capabilities & Differentiators

  • Optimized for Agent Workflows: Designed to reduce ineffective reasoning and accelerate long-running agent tasks, where every-token speed and completion efficiency are paramount.
  • Improved Python Formatting: Addresses and substantially reduces Python indentation issues reported in previous versions, especially with recommended coding parameters (temperature=0.3).
  • Efficiency-Focused Training: Utilizes a fine-tuning philosophy centered on practical inference efficiency, aiming to provide a more practical experience for users with constrained resources.
  • Strong Performance in Agentic Software Engineering: Achieved 100% pass rate (14/14 tasks) on a held-out agentic software-engineering battery, demonstrating robust problem-solving in complex coding scenarios.
  • MMLU-Pro Accuracy: Reports 92.33% MMLU-Pro accuracy under a 64K context with recommended thinking-mode configuration.

Recommended Use Cases

  • Local MTP speculative decoding: Where decoder throughput and draft acceptance are critical.
  • Long-running agent workflows: Involving repeated tool calls and iterative edits.
  • Resource-conscious workloads: Where wall-clock time and generated-token cost are significant operational constraints.

Finetunes
2 models