IQuest-Coder-V1-40B-Instruct
IQuest-Coder-V1-40B-Instruct is a programming-specialized foundation model developed by iquestlab. It boasts approximately 400B parameters and supports a maximum long context window of 128K.
Parameters
400.0B
Context Window
128K
License
https://github.com/openai/gpt-2/blob/master/LICENSE
Release Date
2026-01-01
API Pricing
API pricing for this model is not yet available
Strengths
- ・Extremely large 400B parameters
- ・Long 128K context support
- ・Powerful coding capabilities
Weaknesses
- ・Huge ~80GB file size
- ・High hardware requirements
- ・Closed-source license restrictions
Use Cases
- ・Analyzing large codebases
- ・Complex software development
- ・Advanced programming assistance
Deep Analysis
Architecture
Dense Transformer (40B, 80 layers)
Code-specialized LLM
Context Window
128K tokens
Native long context
Release Date
March 2026
License
Custom
GitHub Stars
1378
IQuestLab/IQuest-Coder-V1
Family
7B/14B/40B/40B-Loop
Multiple size variants
Key Feature
Code-Flow Training Paradigm
Learns from repo evolution patterns
Strengths
- ・SOTA on SWE-Bench Verified, BigCodeBench, LiveCodeBench v6
- ・Code-Flow training from repository evolution patterns
- ・Multiple size variants (7B/14B/40B)
- ・CLI agent integration (ClaudeCode, OpenCode)
- ・HTML/SVG generation capability
- ・128K native context
Weaknesses
- ・Custom license limits some uses
- ・Newer model with less community tooling
- ・Code-specialized, less general-purpose
Competitor Comparison
| Model | Arena | SWE | GPQA | Price |
|---|---|---|---|---|
| DeepSeek Coder V2 | - | - | - | Free (open) |
| CodeLlama 70B | - | - | - | Free (open) |
| Claude Sonnet 4.6 | - | - | - | Paid |
IQuest-Coder-V1-40B is iquestlab's code-specialized LLM family with 7B/14B/40B variants. It introduces a Code-Flow training paradigm that learns from repository evolution patterns, achieving SOTA on major coding benchmarks with CLI agent integration.
Analysis generated: 2026-05-24