DeepSeek R1 671B-A37B on Mac Studio M5 Ultra 512 GB

684.49B (37B active) on 512 GB at 1,200 GB/s. Hardware details · Model details

Yes. DeepSeek R1 671B-A37B fits on Mac Studio M5 Ultra 512 GB at the recommended MLX 4-bit configuration, requiring approximately 385.7 GB at 8K context. Practical context capacity is 160K.
CompatibilityFits
Calculated fit
Yes
Calculated Uses 80-90% of usable memory. Works, with limited headroom.
Recommended quantization
MLX 4-bit
Calculated mlx
Memory required
385.7 GB
Calculated of 435.2 GB usable — 89%
Max practical context
160K
Calculated model supports 160K

Memory budget at 8K context

Model weights ⓘ
372.9 GB Calculated
KV cache ⓘ
0.5 GB Calculated
Runtime overhead ⓘ
12.2 GB Estimated
Total required
385.7 GB Calculated
Headroom ⓘ
49.5 GB Calculated
Fits with limited headroom (89% of usable memory). A longer context will not leave room for much else.
Unified memory: weights, KV cache and the OS share the same 512 GB pool. We assume 85% is addressable by the GPU, which matches a headless machine with the wired-memory limit raised.
Mixture of experts: all 684.49B parameters must be resident in memory even though only ~37B are active per token. Memory follows total parameters; speed follows active parameters.
Performance

Estimated performance · recommended MLX 4-bit

Decode, prefill and TTFT below are estimates for MLX 4-bit. We do not have a comparable MLX 4-bit measurement on this machine.

Estimated decode
—
Estimated MLX 4-bit; calculated range
Estimated prefill
—
Estimated MLX 4-bit; calculated range
Estimated TTFT at 8K
~21.7–45.1 s
Estimated MLX 4-bit; calculated range
Hardware load reference
225 W
Measured machine-level load; not this model run
No throughput figures for this configuration. We do not estimate performance for a model that cannot load.
How the estimate is calculated
  1. Decode: reading 37.0B active parameters at 4.50 bits/weight takes 19.79 ms at 1200 GB/s x 88% achieved efficiency.
  2. MoE routing penalty of 15% applied: expert gathers are less bandwidth-efficient than a dense sweep.
  3. Prefill: 240 TFLOPS (FP16) x 1.69 calibrated against measured prefill on this platform x 22% assumed model-FLOPs utilisation, divided by 2 x 159.1B parameters per token.
  4. Prefill uses 159.1B effective parameters, not the 37B active in decode: a batch of hundreds of tokens routes across most of the expert pool.

  • MoE decode depends on how well the engine batches expert gathers; real results vary more than for dense models.
  • Prefill throughput is highly engine-dependent. Flash attention, batch size and quantized KV all move this number substantially, and for sparse mixture-of-experts models it is the least reliable figure we produce.
  • This is a calculated estimate, not a measurement. It assumes a single request, a short prompt, no speculative decoding and a warm model already resident in memory.

Estimator version estimator_v1. Stored with every estimated row so old estimates can be regenerated when the model improves.

Submit a benchmarkSubmissions stay pending until reviewed.
This model at other quantizations on this machine
QuantizationResidentDiskTotal neededUtilisationFitMax context
MLX 4-bitrecommended372.9 GB372.9 GB385.7 GB89%Fits160K
Q4_K_M398.1 GB398.1 GB411.5 GB95%Borderline160K
Q5_K_M467.0 GB467.0 GB482.6 GB111%Does not fit0
Q8_0690.9 GB690.9 GB713.1 GB164%Does not fit0
Cost of running DeepSeek R1 671B-A37B on this machineUnited States (federal) · C corporation · 8h/day
Monthly economic cost
$118
Calculated after tax
Codex/Claude Code
$100/mo
≈ $100/mo · local is $18 more
Net cash at purchase
$11,350
Calculated VAT not reclaimable
Total over 5 years
$7,102
Calculated after tax, after resale
Cost per USD/1M tokens
—
Calculated

The $100 comparison uses the Codex Pro 5x / Claude Max 5x plans. A fixed planning conversion is used for USD. This compares monthly spend only: subscriptions have usage limits, local hardware has different capabilities and constraints, and taxes or regional pricing may change the charged amount. Prices checked 7 September 2026.

Purchase-price input: $11,350 in United States (federal), tax/VAT excluded · estimated · Apple Store NL (converted catalogue estimate). This is the localized purchase input; the after-tax economic cost is calculated separately below.

Monthly breakdown

Depreciation
7,945 over 5 years, straight-line to a 3,405 residual
$132.41
Electricity
9.9 kWh/month at 0.140/kWh
$1.38
Cost of capital
4.0%/yr on 7,377 average capital employed
$24.59
Monthly cost before tax$158.39
Electricity tax shield
Running costs are deductible business expenses
−$0.29
First-year expensing
§179 (100.0%)
−$39.72
Monthly economic cost after tax$118.37

Three different numbers, deliberately

Cash cost
$11,350
Money that leaves the bank account on day one, net of reclaimable VAT.
Accounting depreciation
$132.41/month
$7,945 written down over 5 years to a $3,405 residual.
After-tax economic cost
$118.37/month
Depreciation plus running costs plus cost of capital, less the tax those deductions save. This is the figure to compare between machines.
Assumptions and sources (verified 2026-09-07)
VAT rate
0% Official spec
VAT recoverable
0% Official spec
Federal C-corporation estimate at the flat 21% rate. Pass-through entities should use the sole-proprietor estimate as a rougher proxy.
Effective deduction rate
21.00% Calculated
Headline marginal rate 21.00%.
Depreciation
5 years, straight-line Assumption
Residual value
$3,405 (30%) Assumption
Extrapolated from the three-year curve.
Electricity
$0.140/kWh Assumption
Average power draw
56 W Calculated
Load 225 W for 20% of powered hours, idle 14 W for the rest — a machine that is on is not generating tokens the whole time.
Investment allowances
§179 100.0% Official spec
Worth $2,383 in total — first-year expensing that replaces later tax depreciation.
Cost of capital
$24.59/month Assumption
Federal planning estimate only, not tax advice. State and local income tax, sales/use tax and incentives are excluded. It assumes 100% business use, a Section 179 election, enough business income to use it, and that the full annual limit remains available.