Kimi-Linear-48B-A3B-Instruct

Kimi family · 49.0B params

Evidence: 2 sourced runs · 1 independent source · first recorded Aug 13, 2026 · last recorded Aug 13, 2026. What these labels mean.

What the Kimi-Linear-48B-A3B-Instruct numbers mean

The fastest verified result on our board is 280.8 output tok/s on an RTX 5090 at IQ4_XS with a 32768-token context. That is about 3.6 ms per generated token — 500 tokens takes roughly 1.8 seconds. Prompt processing on the same record runs at 3733.3 tok/s, so a 2,000-token prompt is read in about 0.5 seconds. Peak VRAM was measured at 25.8 GB. Across the 2 records we hold for this model the range is 90.0–280.8 tok/s on 2 GPUs, because quantisation, context length and the engine build all move the number.

How much VRAM does Kimi-Linear-48B-A3B-Instruct need?

Kimi-Linear-48B-A3B-Instruct has 49.0B parameters. These estimates use the same formula as the VRAM calculator: weights, plus about 15% for a moderate-context KV cache and runtime overhead, plus a small engine floor.

Long contexts and large batches raise the real number — read the smallest card as "possible with a reduced context". Measured throughput is on the benchmark board.

Models with a similar footprint

Ranked by distance from Kimi-Linear-48B-A3B-Instruct's parameter count. Full benchmark board.

All benchmarks Methodology