What it costs to run MiniMax-M3.1-Flash (MoE)

How these numbers are produced

• Model shape comes from the model's own config.json — 60 layers, 64 attention heads, 4 KV heads.
• Weight size is parameters × effective bits-per-weight. Those constants are checked against real published quantised file sizes — median error 0.7% across 40 measurements.
• KV cache is 2 × layers × KV-heads × head-dim × context × 2 bytes. Using KV-heads rather than attention heads is what makes this correct for grouped-query attention; treating a GQA model as multi-head overstates the cache by up to 8×.
• Tokens/sec is an estimate, not a benchmark. Generation is memory-bandwidth-bound, so this is bandwidth ÷ weight-bytes derated to 75%. Real throughput depends on your runtime, batch size and kernels. We don't run our own hardware tests — see our editorial policy.
• This is a mixture-of-experts model. Generation reads only the experts active for each token (4 of 128), so throughput is scored on ~23.0B active parameters, not the full 427B. Memory is the opposite: every expert must still be resident in VRAM, so the figures above use the full weight set.MoE: first 3 layers dense, 57 MoE layers with 128 routed experts (top-4) + 1 shared. All 60 layers use MiniMax Sparse Attention (block-sparse, GQA 4 KV heads x 128), which picks blocks to read but caches every token, so all 60 count; the indexer adds one 128-dim key per token per layer (+12.5%, not counted). params = 427.05B counted from tensor shapes, NOT the HF API figure (220.5B), which counts the NVFP4-packed experts (2 weights per byte) once and skips scale tensors; the official QAT checkpoint is ~250 GB (NVFP4 experts, MXFP8 attention) and excludes the 1.96B DSpark draft head. Card: ~428B total / 23B active, up to 1M context. HF config + card read Nov 7, 2026.
• Some memory-bandwidth figures are not yet independently verified.Where that's the case we show no tokens/sec at all rather than a number we can't stand behind.
• Rental prices are pulled hourly from provider APIs (last updated Sat, 07 Nov 2026 02:20:13 GMT). Spot/interruptible pricing can change or vanish without notice.