1× Mac
155tok/sEveryday
1 request, llmprobe standard prompt, thinking on, with MTP + PLD (per recipe) Source
Intelligence40with thinking
- Experimental
- License unclear
Local AI · Recipe · 1× Mac
Gemma-4 MLX 4-bit with MLX on 1× Mac
Gemma-4 MLX 4-bit with MLX on 1× Mac: 102 tok/s according to github.com (dataset as of Nov 3, 2026).
by 0xBakeer
Intelligence (original model) · Artificial Analysis8with thinkingno thinking 7
Engine
- Engine
- LM Studio (MLX-Backend) 0.4.21
- Quantization
- mlx-4bit
- Model family
- Gemma-4
- Context
- 131,072
- Parameters
- E2B (Größe lt. Modellname)
- Creator
- 0xBakeer
- GitHub stars
- 16
- Repo updated
- Oct 19, 2026
Measurements
Every sourced value of this recipe, each with its condition and source. Bars relative to the largest value in the group.
Decode · 1 request
Prefill by context
Time to first token
Shorter is better.
What you need
- Hardware
- 1 × Mac / Apple Silicon
- Weights
- Main weightsgoogle/gemma-4-E2B-it Apache-2.0
- Engine
- LM Studio (MLX-Backend) 0.4.21
- Context
- 131,072 tokens
Sources
- github.com/0xBakeer/inference-atlas/blob/main/results/lmstudio/google/gemma-4-E2B-it/apple-m2-max-32gb/251c7884bb174170--serve-single-i256-o256-v1--57e31f.json
- github.com/0xBakeer/inference-atlas/blob/main/results/lmstudio/google/gemma-4-E2B-it/apple-m2-max-32gb/251c7884bb174170--prefill-8k-v1--d72bb6.json
- github.com/0xBakeer/inference-atlas/blob/main/results/lmstudio/google/gemma-4-E2B-it/apple-m2-max-32gb/251c7884bb174170--serve-chat-c32-i1k-o256-v1--a9e372.json
- github.com/0xBakeer/inference-atlas/blob/main/hardware/apple-m2-max-32gb.json
Related recipes
1× Mac
53.3tok/sEveryday
1 request, prose prompt, with MTP (nativ) (per recipe) Source
Intelligenceno independent value
- License unclear
1× Mac
35.8tok/sEveryday
1 request, llmprobe standard prompt, thinking on, with embedded MTP (2 draft tokens) (per recipe) Source
Intelligence42with thinking
- Experimental
- No license on HF
- License unclear
1× Mac
28.5tok/sEveryday
1 request, prose prompt, context 32k, with DSpark (per recipe) Source
Intelligenceno independent value
- ≤ 3 bit: quantization may cost quality
1× Mac
25.1tok/sEveryday
1 request, prose prompt, short prompt, with MTP (DSpark built-in) (per recipe) Source
Intelligence40with thinkingno thinking 25
- ≤ 3 bit: quantization may cost quality
- No license on HF
- License unclear
1× MacBonsai-2-27B (Qwen3.8-27B-Basis)
10.6tok/sEveryday
1 request, prose prompt, no speculative decoding Source
Intelligenceno independent value
- Custom kernel required
Local AI in your company?
In a workshop we work out which models and which hardware fit your tasks, and build the first agent on your infrastructure.