Is iQuest Coder really better than Claude Sonnet 4.5?

In specific coding benchmarks like SWE-Bench Verified, iQuest Coder scored 81.4%, which is higher than Sonnet 4.5's reported scores. However, some independent tests suggest the score may be inflated due to data contamination, with corrected scores sitting closer to 76%, which is still highly competitive.

What hardware do I need to run it?

The 40B parameter model is designed to fit on a single high-end consumer GPU, such as an NVIDIA RTX 3090 (24GB) or RTX 4090. Smaller 7B and 14B versions are available for laptops.

Who created this model?

It was created by Quest Research, a lab affiliated with Ubiquant, a prominent Chinese quantitative hedge fund. This explains the model's focus on logic and complex algorithmic efficiency.

What is "Loop" architecture?

Unlike standard models that process data in one straight line, the "Loop" model passes data through its neural network layers twice. This allows it to "re-read" and refine its own output before finishing, improving accuracy on complex coding logic without needing a larger model size

IQuest Coder V1

The open-source “Loop” model that beats the giants.

From Ubiquant

Founder: Quest Research

Text Generator

1 Jan 2026

20K+

Free

Pricing Model

Free

Starting Price

$0/Month

About This AI

iQuest Coder is a 40-billion parameter open-source AI model designed for coding.

It uses a unique “Loop” architecture and “Code-Flow” training (learning from git commits) to achieve state-of-the-art performance, reportedly beating Claude Sonnet 4.5 and Gemini 3.0 on benchmarks.

Chatbot Coding Conversation LLM

Pricing

Pricing Model

Free

Starting Price

$0/Month

Key Features

Loop Architecture: Runs layers twice for recursive reasoning.
Code-Flow Training: Learns from git commit history and project evolution.
High Context: Supports up to 128K context window.
Local Run: Optimized for RTX 3090/4090 (40B parameters).

Pros

Beats top proprietary models (Sonnet 4.5, Gemini 3.0) on SWE-Bench.
Open-source and free to use locally.
Highly efficient "Loop" design saves memory.

Cons

Initial benchmark scores faced scrutiny regarding git history data leakage (corrected score ~76.2%).
Slightly slower inference due to "Loop" (double pass) architecture.

Best For Local software development, complex algorithmic problem solving, and automated bug fixing.

FAQs

Is iQuest Coder really better than Claude Sonnet 4.5?
In specific coding benchmarks like SWE-Bench Verified, iQuest Coder scored 81.4%, which is higher than Sonnet 4.5’s reported scores. However, some independent tests suggest the score may be inflated due to data contamination, with corrected scores sitting closer to 76%, which is still highly competitive.
What hardware do I need to run it?
The 40B parameter model is designed to fit on a single high-end consumer GPU, such as an NVIDIA RTX 3090 (24GB) or RTX 4090. Smaller 7B and 14B versions are available for laptops.
Who created this model?
It was created by Quest Research, a lab affiliated with Ubiquant, a prominent Chinese quantitative hedge fund. This explains the model’s focus on logic and complex algorithmic efficiency.
What is “Loop” architecture?
Unlike standard models that process data in one straight line, the “Loop” model passes data through its neural network layers twice. This allows it to “re-read” and refine its own output before finishing, improving accuracy on complex coding logic without needing a larger model size