Benchmarks & data

Raw, sourced benchmark data with full methodology — tok/s, memory, accuracy — plus downloadable CC-BY datasets you can cite.

Benchmark data

Dataset

Can a local 27B model code like Claude?

The measured reverse-engineering: ~45% on SWE-bench Verified (18/40 blind), mid-50s to low-60s with local engineering, and a genuine 15–20-point wall. Where it passed tests, the 3 hard walls, and the biggest lever.

Dataset

Local AI benchmarks for Mac (2026)

Tok/s and memory data for local AI on Apple Silicon in 2026: Outlier measured, Ollama and LM Studio from public benchmarks.

Dataset

Mac RAM to AI model size — reference table

A reference table mapping Mac unified RAM to which local AI models fit, their disk sizes, and measured tokens/sec on an M1 Ultra. Plan your build or buy.

Dataset

Outlier vs Claude: 54-prompt benchmark

Raw side-by-side data: 54 prompts answered by Outlier Core 27B and Claude Opus 4.7, scored against a rubric we wrote ourselves. Compare the results yourself.

Dataset

Outlier tier benchmarks — MMLU for every shipping tier

MMLU for every shipping tier, measured through the shipped 1.11.804 build on an M1 Ultra: Nano 52.0%, Lite 77.0%, Quick 81.0%, Core 89.5%, Vision 3.8 82.5%. n=200, with the superseded 1.11.757 run kept for comparison.

Dataset

Local AI time to first token on a Mac

How long a new chat waits for its first word, before and after the system prompt's work is saved and reused. Core and Vision 3.8: about 20 s to about 1 s, with the same words. Raw tables from an M1 Ultra.

Dataset

Local AI math word problems on a Mac

Outlier Lite and Nano on 20 short word problems: Lite 12 right answering straight away and 20 after working each one out first, Nano 10 and then 60 of 60 over three runs. The cost is 6 to 8 s instead of 0.4 s. Raw tables from an M1 Ultra.

Dataset

Why a local AI gives the same answer every time

Outlier Nano, Lite and Core gave the identical reply 3 of 3 times on every test at temperature 0.7. The cause, the fix, and a 2-minute test for your own app. Tables from an M1 Ultra.

Dataset

Streaming engine benchmarks (Plus + Vision)

Raw tok/s and memory data for Outlier's MoE engines on Plus 397B and Vision 35B. Why both streaming engines (V10, V11) were retired in favor of V9…

Try Outlier free

Free: Nano + Lite. Pro: $249 once · or 4 × $62.25. macOS 26+. In the US, Klarna or Afterpay at checkout: four payments, two weeks apart. Refund window: 30 days, no questions asked.

Download free (290 MB) Buy Pro

New models worth running