Local AI Advisor

Can Ryzen AI Max+ 395 mini PC / Framework Desktop · 64 GB run Kolibri-1 (78B-A3.5B)?

64 GB unified · 256 GB/s memory bandwidth · Kolibri-1 (78B-A3.5B) has 78.1B parameters (3.46B active per token).

Short answer

Yes, excellent for chat; borderline for agentic coding.

Best quantization for chat: Q4_K_M (45 GB of weights) on Lemonade. Generation: 65–90 tok/s, leaving about 6.8 GB free for your other apps.

How it rates for each workload

Same model and machine, different jobs. Agents and long documents send far bigger prompts than chat, so they need much more speed and memory.

WorkloadVerdictQuantGenerationFirst replyContext
Chatwith Open WebUIExcellentQ4_K_M65–90 tok/s≈1 s8KDetails
Coding questionswith ContinueComfortableQ4_K_M65–90 tok/s≈6 s16KDetails
Coding in a repositorywith AiderAcceptableQ4_K_M60–80 tok/s≈15 s32KDetails
Agentic codingwith OpenCodeBorderlineQ4_K_M55–75 tok/s≈25 s32KDetails
Long documents (64K)with Open WebUIAcceptableQ4_K_M45–60 tok/s≈65 s64KDetails

Speeds are estimates calibrated against measured benchmarks on similar hardware. “First reply” is the time to read the workload’s opening prompt. How we calculate

Different tool, context size or workload?

These pages use typical settings. The full check models your exact setup.