Evidence: benchmark results vendor-reported
Leanstral 1.5 119B-A6B — VRAM, speed & local setup
Written by Jakub Rusinowski · Last updated
A Lean 4 proof agent: 119B parameters with 6B active per token (128 experts, 4 active) and a 256K context, of which Mistral recommends using at most 200K. It writes and repairs formal proofs; it is not a general chat model. Apache-2.0, FP8 weights on Hugging Face.
Leanstral 1.5 119B-A6B needs about 73 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.
VRAM and speed by quantization
Quoted against NVIDIA RTX 4090 (24 GB). Includes the KV cache at 8K context, so it reads higher than the headline figure.
| Quant | Memory | VRAM | Speed (est.) | Fit |
|---|---|---|---|---|
| Q2_K 2.63 bpw | 40.1 GB | ~30 tok/s | Offload | |
| Q3_K_M 3.41 bpw | 51.7 GB | ~26 tok/s | Offload | |
| Q4_K_M 4.83 bpw | 72.8 GB | — | Too big | |
| Q5_K_M 5.67 bpw | 85.3 GB | — | Too big | |
| Q6_K 6.56 bpw | 98.6 GB | — | Too big | |
| Q8_0 8.50 bpw | 127.4 GB | — | Too big | |
| F16 16.00 bpw | 239 GB | — | Too big |
Black marker = usable memory on the NVIDIA RTX 4090 (24 GB). Estimates from the memory-bandwidth roofline on the methodology page. Leanstral 1.5 119B-A6B VRAM calculator →
Get Leanstral 1.5 119B-A6B running
The cheapest catalogued GPU that runs Leanstral 1.5 119B-A6B is the NVIDIA DGX Spark (128 GB).
or compare on Vast.ai from $0.77/hr (typical low · varies)
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
How to run Leanstral 1.5 119B-A6B
No Ollama tag. Mistral's card documents vLLM; its hosted API was retired on 30 September 2026.
Specifications
Verified — Checked against the primary source — the model card or the vendor spec page — and corroborated by a second independent source.
- Parameters
- 119 Billion (6B active)
- Context window
- 262,144
- Architecture
- Mixture-of-Experts (128 experts, 4 active), MLA
- Provider
- Mistral AI
- Licence
- Apache-2.0
- Practical deployment
- Workstation (large unified memory or pro GPU)
- Specified at
- Q4_K_M
- System RAM
- 128 GB
- Record updated
- 2026-11-03
Commercial use permitted. No usage restrictions beyond attribution.
Limits and caveats
- Mistral retired Leanstral's hosted API on 30 September 2026; the open weights are now the only way to use it.
- A Lean 4 specialist, kept out of general coding rankings on purpose.
See also Mistral Small 4 119B-A6.5B
Quality and use cases
Scores as published by the model’s authors or an independent evaluator — quality, not throughput, and not measured by us.
Leanstral 1.5 119B-A6B — frequently asked questions
How much VRAM does Leanstral 1.5 119B-A6B need?
About 73 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.
Does Leanstral 1.5 119B-A6B run on an RTX 4090 (24 GB)?
No. Leanstral 1.5 119B-A6B needs about 73 GB at Q4_K_M, more than a single RTX 4090's 24 GB. It needs a larger card, several GPUs, or Apple Silicon with enough unified memory — or it runs with part of the weights offloaded to system RAM, which is much slower.
How do I run Leanstral 1.5 119B-A6B locally?
No Ollama tag. Mistral's card documents vLLM; its hosted API was retired on 30 September 2026. Running the published tag would send your prompts to a hosted GPU rather than your own machine.