roskosmos19/Rhea-4B-fast-0409

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 6, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Rhea-4B-fast-0409 by roskosmos19 is a 4 billion parameter Qwen3ForCausalLM-based model with a 32,768 token context window, specifically optimized for agentic workflows and coding tasks. This iteration focuses on faster, cheaper inference by streamlining its reasoning style to a single-pass approach with optional thinking, departing from multi-pass generation. It retains strong coding and reasoning capabilities while significantly reducing inference cost and latency compared to its predecessors.

Loading preview...

Overview

roskosmos19's Rhea-4B-fast-0409 is a 4 billion parameter model built on the Qwen3ForCausalLM architecture, featuring a 32,768 token context length. It represents an optimized successor in the Rhea-4B-Coding / Athenea lineage, engineered for enhanced performance and cost-efficiency in agentic and coding applications. The model's core improvement lies in its shift from a forced multi-pass reasoning style to a more flexible single-pass approach, incorporating an optional <think>...</think> mechanism for chain-of-thought processes.

Key Optimizations & Features

  • Streamlined Reasoning: Moves from a forced three-pass reasoning (implement→review→final) to a single-pass method, reducing latency and inference costs.
  • Optimized Context Handling: While reducing the maximum context from 262k to 32,768 tokens, it is noted that this length adequately covers real-world agentic workloads.
  • Cost-Efficiency: Designed for significantly lower inference cost and VRAM usage due to the removal of forced long outputs and multi-pass overhead.
  • Clean Special Tokens: Features a lean set of special tokens focused on tools and thinking, avoiding broken prefixes and numerous vision tokens.
  • Improved Agentic Readiness: Incorporates an enhanced tool-calling template and standard <think> tokens for better agentic behavior.
  • Strong Coding & Reasoning: Retains its focus on robust coding and reasoning capabilities, encouraging precise, secure, and efficient solutions via its system prompt.

Recommended Use Cases

  • Agentic Workflows: Ideal for applications requiring efficient, single-pass agentic reasoning and tool use.
  • Code Generation & Analysis: Suited for tasks involving secure Python function generation, JWT validation, and general coding challenges.
  • Cost-Sensitive Deployments: Recommended for scenarios where inference cost and speed are critical, especially when deployed with optimizations like Q4_K_M / AWQ quantizations or vLLM/SGLang.