Last Updated: April 2026
Table of Contents¶
- Hugging Face Inference Providers
- OpenCode Free Models
- Other Free Tiers
- Recommendations by Use Case
- Model Comparison Tables
Free AI Models Overview¶
Hugging Face Inference Providers¶
OpenCode natively supports Hugging Face Inference Providers - giving you access to open models from 17+ providers.
Quick Setup¶
Hugging Face Setup
```bash
1. Create token at huggingface.co/settings/tokens¶
(needs "Make calls to Inference Providers" permission)¶
2. Run auth login¶
opencode auth login
3. Select Hugging Face when prompted¶
Enter your token: hf_...¶
4. Select a model¶
/models
### Best Coding Models
| Model | Best For | Provider | Context | Price (Input/Output) |
| --------------------- | ----------------- | ----------- | ------- | -------------------- |
| **Qwen3-Coder-Next** | Code generation | Novita | 131K | $0.10/$1.50 |
| **Qwen2.5-Coder-32B** | Code reasoning | Featherless | 131K | Varies |
| **DeepSeek-R1** | Complex reasoning | Hyperbolic | 131K | Free tier |
| **GLM-4.7 Flash** | Fast/short tasks | Zai | 1M | Free tier |
### Top Recommendation: Qwen3-Coder-Next
**Why it's like Claude/Big Pickle:**
- Code-specific training (trained specifically for coding)
- Strong on HumanEval, MBPP+ benchmarks
- Fast inference (128 tokens/sec via Novita)
- Excellent for code completion and generation
**Benchmark Results:**
| Model | HumanEval | HumanEval+ | MBPP+ |
| ----------------- | --------- | ---------- | ----- |
| Qwen3-Coder-Next | SOTA | SOTA | SOTA |
| Qwen2.5-Coder-32B | 76.2% | 72.6% | 70.9% |
### Other Notable Models
**For Reasoning:**
- DeepSeek-R1 - Chain-of-thought reasoning, great for complex logic
- Kimi-K2.5 - Fast, good for general tasks
- GLM-4.6 - Free tier available
**For General Use:**
- Llama 3.1 8B - Fast, cheapest
- Qwen2.5 7B - Good balance
- Gemma 4 31B - Google's best
**Resources:**
- [HF Supported Models](https://huggingface.co/inference/models)
- [HF Inference Providers](https://huggingface.co/docs/inference-providers/en/index)
---
## OpenCode Free Models
OpenCode offers some built-in free models developed by [Zen Labs](https://zenlabs.ai/):
| Model | Best For | Notes |
| :--- | :--- | :--- |
| **Big Pickle** | General coding | OpenCode's free tier - developed by Zen |
| **GLM 4.7 Free** | General tasks | Limited |
| **MiniMax M2.1 Free** | General tasks | Limited |
### About Big Pickle
**Big Pickle** is a free LLM developed by [Zen Labs](https://zenlabs.ai/), the team behind OpenCode. It's available directly in OpenCode and optimized for coding tasks.
<div class="youtube-video-wrapper">
<iframe src="https://www.youtube.com/embed/tuW0IKNZ2UI" allowfullscreen></iframe>
</div>
### How to Access
```bash
opencode
# In terminal:
/models
# Select from available free models
Other Free Tiers¶
Free Tier Comparison¶
| Provider | Free Credits | Best Models | Sign Up |
|---|---|---|---|
| Google AI Studio | 15 RPM, 250K TPM | Gemini 2.5 Pro/Flash | aistudio.google.com |
| GitHub Models | 50-150 req/day | GPT-4.1, o3 | github.com/models |
| NVIDIA NIM | 1,000 credits | DeepSeek R1, Llama | build.nvidia.com |
| Hugging Face | Monthly credits | 300+ models | huggingface.co |
| Groq | Limited | Llama, Qwen | console.groq.com |
| xAI | $25 credits | Grok 4 | x.ai |
Google AI Studio (Recommended for Free)¶
Best Free Option
- Gemini 2.5 Pro (best free model)
- Gemini 2.5 Flash (fastest)
- 1M token context
- Sign up at aistudio.google.com
GitHub Models¶
- GPT-4.1 (excellent coding)
- o3 (reasoning)
- Direct integration with OpenCode via /connect
Resources:
Recommendations by Use Case¶
Best for Coding Tasks¶
- Qwen3-Coder-Next (HF Novita) - Code-specific, fast
- GPT-4.1 (GitHub Models) - Best overall coding
- Gemini 2.5 Pro (Google AI Studio) - Long context
Best for Reasoning¶
- DeepSeek-R1 (HF Hyperbolic) - Chain-of-thought
- o3 (GitHub Models) - Complex logic
- Gemini 2.5 Pro (Google) - Long context
Best for Speed¶
- Groq - Fastest inference (1000+ t/s)
- Cerebras - Ultra-fast
- Qwen3-Coder-Next - 128 t/s
Best for Free Tier (No Credit Card)¶
- Big Pickle (OpenCode built-in)
- GLM-4.7 Flash (HF)
- Gemini 2.5 Flash (Google)
Model Comparison Tables¶
Coding Models Comparison¶
| Model | Provider | Context | Speed | Price | Best For |
|---|---|---|---|---|---|
| Qwen3-Coder-Next | Novita | 131K | 128 t/s | \(0.10/\)1.5M | Code generation |
| Qwen2.5-Coder-32B | Featherless | 131K | Medium | \(0.50/\)1M | Code reasoning |
| DeepSeek-R1 | Hyperbolic | 131K | Medium | Free | Complex reasoning |
| GPT-4.1 | GitHub | 32K | Fast | Free | General coding |
Free Models Comparison¶
| Model | Source | Free Tier | Notes |
|---|---|---|---|
| Big Pickle | OpenCode | Unlimited | Works out of box |
| GLM 4.7 Flash | Hugging Face | Yes | Slower |
| Gemini 2.5 Flash | Generous | Best value | |
| Gemma 4 31B | Hugging Face | Limited | Google quality |
TODO: Review Schedule¶
- Review monthly - check for new models
- Check free tier limits changed
- Update recommendations based on benchmarks
Resources¶
This is a living document. Revisit and update regularly.