whichllm

About whichllm

CLI tool that finds the local LLM that best fits a specific hardware setup, ranked by real recency-aware benchmark results rather than marketing claims.

Best for: Developers and builders who want a hardware-aware recommendation of which local model to run, based on actual benchmark performance on their GPU and RAM configuration.

Deployment: CLI tool -- macOS, Linux, Windows via terminal

Skill level: Beginner to intermediate

Tradeoffs

Benchmark coverage depends on upstream data freshness -- models released very recently may have limited benchmark signal. Recommendations are performance-focused, not qualitative -- treat output as a starting shortlist, not a final answer.

Related guides and resources

Explore step-by-step setup guides, comparisons, and stack recipes for this tool category.

Best for

Developers and builders who want a hardware-aware recommendation of which local model to run, based on actual benchmark performance on their GPU and RAM configuration.

Why use it

whichllm closes the gap between compatibility (will a model fit?) and performance (which model actually runs best?) -- a distinction the compatibility checker and model browser do not fully address. It pulls benchmark data, weights it for recency, and outputs a ranked recommendation for a specific hardware profile.

Key features

• Hardware-aware model ranking: outputs recommendations filtered by VRAM, RAM, and CPU profile rather than generic benchmark leaderboards
• Recency-weighted benchmark data: recent benchmark results carry more weight than older scores, reducing stale recommendation bias
• Apple Silicon support: M-series unified memory profiles are treated correctly alongside discrete GPU VRAM configurations
• CLI output: clean ranked list of models with benchmark scores, VRAM fit, and context window notes

Common AI use cases

• Get a ranked shortlist of models before pulling from Ollama or LM Studio
• Compare benchmark-weighted performance across quantization levels on a given GPU
• Identify the best-fit model for Apple Silicon unified memory configurations

Who should use it

• Builders who want benchmark signal -- not just VRAM fit -- before committing to a model download
• Apple Silicon users who want M-series-aware model recommendations
• Engineers evaluating local model performance before deploying a local AI feature

Who should not use it

• Users who only need VRAM compatibility -- the OpenSourcesAI checker is simpler for that use case
• Teams requiring qualitative evaluation (instruction following, coding quality) -- pair with manual testing

Tradeoffs

Benchmark coverage depends on upstream data freshness -- models released very recently may have limited benchmark signal. Recommendations are performance-focused, not qualitative -- treat output as a starting shortlist, not a final answer.

Alternatives

• Ollama
• LM Studio