You have 8GB RAM and want a local coding model. Three contenders: Yi-Coder 9B (purpose-built for code), Qwen3 8B (best all-rounder), and Falcon H1R 7B (best reasoning). Here’s which to pick.

Head-to-head

	Yi-Coder 9B	Qwen3 8B	Falcon H1R 7B
Parameters	9B	8B	7B
RAM needed	8 GB	8 GB	6 GB
Architecture	Dense	Dense	Hybrid SSM + attention
Coding focus	✅ Purpose-built	General + coding	General + reasoning
Languages	52	30+	20+
Context	128K	32K	8K
FIM (autocomplete)	✅	✅	❌
Reasoning	Decent	Good	✅ Best
License	Apache 2.0	Apache 2.0	Apache 2.0

For each coding task

Task	Best pick	Why
Code generation	Yi-Coder 9B	Trained specifically on code
Code autocomplete	Yi-Coder 9B	FIM support, 52 languages
Debugging	Falcon H1R 7B	Better reasoning finds root causes
Code review	Qwen3 8B	Best balance of code + language understanding
Refactoring	Yi-Coder 9B	Understands code structure best
Documentation	Qwen3 8B	Best writing quality
Algorithm design	Falcon H1R 7B	Strongest math/logic
Quick questions	Any	All fast at this size

Speed comparison

On a MacBook Air M2 16GB:

Model	Tokens/second	Time for 500-token response
Falcon H1R 7B	~28 tok/s	~18 seconds
Qwen3 8B	~25 tok/s	~20 seconds
Yi-Coder 9B	~22 tok/s	~23 seconds

Falcon H1R is fastest due to its smaller size and hybrid architecture. All three are fast enough for interactive coding.

Setup

# All three via Ollama
ollama pull yi-coder:9b
ollama pull qwen3:8b
ollama pull falcon2  # or falcon-h1r when available

# Test each
ollama run yi-coder:9b "Write a REST API with authentication in Express.js"
ollama run qwen3:8b "Write a REST API with authentication in Express.js"

Connect to Aider:

aider --model ollama/yi-coder:9b    # Best for coding
aider --model ollama/qwen3:8b       # Best all-rounder

The practical recommendation

If you only install one: Qwen3 8B — best all-rounder, handles coding + chat + reasoning well.

If coding is all you do: Yi-Coder 9B — purpose-built, 52 languages, FIM for autocomplete.

If you need reasoning: Falcon H1R 7B — hybrid architecture beats both on logic tasks.

If you have 16GB+ RAM: Skip all three and use Devstral Small 24B — significantly better quality.

Real-world performance notes

These benchmarks tell part of the story. In practice:

Yi-Coder 9B shines on multi-language projects. If your codebase mixes Python, TypeScript, and Go, Yi-Coder handles the context switching better than Qwen3 8B because it was trained on 52 languages specifically. It also has the longest context window (128K vs 32K for Qwen3), which matters when you need to feed multiple files into context.

Qwen3 8B is the safest default. It handles coding, chat, reasoning, and documentation all at a “good” level. If you only install one model, this is it. The trade-off is it doesn’t excel at any single task the way Yi-Coder excels at code or Falcon H1R excels at reasoning.

Falcon H1R 7B is the surprise performer. Its hybrid Transformer-Mamba architecture gives it 256K context and 2x the throughput of Qwen3-8B (1,500 tok/s vs ~750 tok/s per GPU at batch size 64). For reasoning-heavy coding tasks (algorithm design, debugging complex logic, mathematical code), it outperforms models up to 7x its size including Phi-4 14B and Qwen3 32B.

The two-model setup

If you have the RAM, run two models:

# Yi-Coder for code generation and editing
aider --model ollama/yi-coder:9b

# Falcon H1R for debugging and reasoning
# Switch when you hit a hard problem
aider --model ollama/falcon-h1r:7b

This gives you the best of both worlds: Yi-Coder’s code expertise for 80% of tasks, Falcon H1R’s reasoning for the hard 20%. Total cost: $0.

When none of these are enough

At the sub-10B level, you’re trading quality for speed and RAM efficiency. For tasks where quality matters most:

Need	Upgrade to	RAM needed
Better coding	Devstral Small 24B	16 GB
Better reasoning	DeepSeek R1 14B	12 GB
Best all-rounder	Qwen 3.5 27B	20 GB
Frontier quality	Claude Code ($20/mo)	API
Free frontier	Qwen 3.6 Plus	API (free)

Related: Yi-Coder Guide · Best Ollama Models for Coding · How to Run Falcon Locally · How to Run Yi Locally · VRAM Guide · Free AI Coding Server