OpenChat 3.5 7B (Free)

Overview

OpenChat 3.5 7B is a high-performance open-source language model fine-tuned from Mistral-7B-v0.1 using C-RLFT (Conditioned Reinforcement Learning Fine-Tuning). It represents one of the best-performing open-source 7B models, outperforming many larger models on various benchmarks.

Description

OpenChat is a library of open-source language models fine-tuned with C-RLFT, a strategy inspired by offline reinforcement learning. The model is trained on mixed-quality data without preference labels, achieving remarkable performance that rivals or exceeds many proprietary models.

Key innovations:

• C-RLFT Training: Conditioned Reinforcement Learning Fine-Tuning methodology
• Mixed-Quality Data: Training without requiring preference labels
• Research-Backed: Based on research paper arXiv:2309.11235

Technical Specifications

Specification	Value
Context Length	8,192 tokens
Input Modalities	Text
Output Modalities	Text
Tensor Type	BF16
Format	Safetensors
Base Model Group	Mistral
Instruction Type	OpenChat
Default Stop Sequences	`<
Minimum GPU RAM	24GB

Capabilities

Capability	Supported
Chat/Conversation	Yes
Code Generation	Yes (Strong)
Mathematical Reasoning	Yes (Specialized Mode)
General Knowledge	Yes
Instruction Following	Yes
Multi-turn Conversations	Yes
Function Calling	No
Vision/Image Input	No
Reasoning Traces	No

Supported Parameters

Parameter	Description	Recommended Range
temperature	Controls randomness	0.0 - 2.0
top_p	Nucleus sampling	0.0 - 1.0
top_k	Top-k sampling	1 - 100
max_tokens	Maximum output tokens	1 - 8192
stop	Stop sequences	`["<
frequency_penalty	Reduce repetition	-2.0 - 2.0
presence_penalty	Encourage diversity	-2.0 - 2.0

Related Models

Model	Description
openchat/openchat-7b	Base OpenChat 7B (fine-tuned on Mistral 7B)
openchat/openchat-8b	Newer version fine-tuned on Llama 8B
mistralai/mistral-7b-instruct	Base Mistral model

Model Information

Property	Value
Model Name	OpenChat 3.5 7B
Model ID	openchat/openchat-7b:free
HuggingFace Model	openchat/openchat-3.5-0106
Author	OpenChat
Parameter Count	7 Billion
Base Model	Mistral-7B-v0.1
Created	November 28, 2023
Last Updated	November 10, 2025
License	Apache-2.0

Pricing (OpenRouter Free Tier)

Type	Price
Input	Free ($0/1M tokens)
Output	Free ($0/1M tokens)

Note: Free tier models on LangMart may have rate limits and usage restrictions.

Performance Benchmarks

Overall Performance Comparison

Model	Avg Score	MT-Bench	HumanEval	BBH MC	AGIEval	TruthfulQA	MMLU	GSM8K	BBH CoT
OpenChat-3.5-0106	64.5	7.8	71.3	51.5	49.1	61.0	65.8	77.4	62.2
ChatGPT (March)	61.5	7.94	48.1	47.6	47.1	57.7	67.3	74.9	70.1
OpenChat-3.5-1210	63.8	7.76	68.9	49.5	48.0	61.8	65.3	77.3	61.8

HumanEval+ (Coding Benchmark)

Model	Pass@1	Parameters
OpenChat-3.5-0106	65.9%	7B
ChatGPT (Dec 2023)	64.6%	Unknown
WizardCoder-Python-34B	64.6%	34B

Key Performance Highlights

• Outperforms Grok-0 (33B) on all 4 tested benchmarks
• Outperforms Grok-1 on average and 3 out of 4 benchmarks
• 15-point improvement in coding over previous OpenChat-3.5 version
• Best performing open-source 7B model in its class

Operating Modes

1. Default Mode (GPT4 Correct)

Best for coding, chat, and general tasks.

Conversation Template:

GPT4 Correct User: [user_message]<|end_of_turn|>GPT4 Correct Assistant: [assistant_response]<|end_of_turn|>

2. Mathematical Reasoning Mode

Tailored for solving math problems.

Conversation Template:

Math Correct User: [math_problem]<|end_of_turn|>Math Correct Assistant: [solution]<|end_of_turn|>

Usage Examples

LangMart API Request

curl https://api.langmart.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $LANGMART_API_KEY" \
  -d '{
    "model": "openchat/openchat-7b:free",
    "messages": [
      {"role": "user", "content": "Explain the concept of machine learning in simple terms."}
    ]
  }'

Python with OpenAI SDK

from openai import OpenAI

client = OpenAI(
    base_url="https://api.langmart.ai/v1",
    api_key="your-langmart-api-key"
)

response = client.chat.completions.create(
    model="openchat/openchat-7b:free",
    messages=[
        {"role": "user", "content": "Write a Python function to calculate fibonacci numbers."}
    ]
)

print(response.choices[0].message.content)

Self-Hosted Deployment

python -m ochat.serving.openai_api_server \
  --model openchat/openchat-3.5-0106 \
  --engine-use-ray \
  --worker-use-ray

Server runs at localhost:18888 with OpenAI-compatible API.

Known Limitations

• Complex multi-step reasoning tasks may be challenging
• May hallucinate non-existent information
• Potential for generating harmful or biased outputs
• Some mathematical and arithmetic problems may produce incorrect results
• Limited context window (8K tokens) compared to newer models

Citation

@article{wang2023openchat,
  title={OpenChat: Advancing Open-source Language Models with Mixed-Quality Data},
  author={Wang, Guan and Cheng, Sijie and Zhan, Xianyuan and Li, Xiangang and Song, Sen and Liu, Yang},
  journal={arXiv preprint arXiv:2309.11235},
  year={2023}
}

Resources

• LangMart Model Page
• HuggingFace Model Repository
• Research Paper (arXiv)
• GitHub Repository

---

Last Updated: December 2025