Recursive Flint 40B

Flint family · 40.0B params

2recorded runs

224.8best tok/s

2verified

Evidence: 2 sourced runs · 1 independent source · first recorded Oct 23, 2026. What these labels mean.

What the Recursive Flint 40B numbers mean

The fastest verified result on our board is 224.8 output tok/s on an RTX 5090 at NVFP4 with a 262144-token context. That is about 4.4 ms per generated token — 500 tokens takes roughly 2.2 seconds.

How much VRAM does Recursive Flint 40B need?

Recursive Flint 40B has 40.0B 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.

Quantisation	Estimated VRAM	Smallest card that fits
Q4_K_M (4-bit)	26.7 GB	32 GB
Q8_0 (8-bit)	49.7 GB	64 GB
FP16 (16-bit)	92.8 GB	141 GB

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.

Fastest GPUs for this model (best tok/s)

RTX 5090224.8

Models with a similar footprint

Model	Params	Q4_K_M weights
Ornith-1.5-35B-A3B	36B	20.2 GB
Ornith-1.5-35B-A3B-GPTQ-Int4-sym-G128-MTP-BF16-MixedCal-v2	36B	20.2 GB
Ornith-1.5-35B-A3B-MTP	36B	20.2 GB
Qwen3.6-35B-A3B	36B	20.2 GB
Qwen3.6-35B-A3B-abliterated-v4	36B	20.2 GB
Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved	36B	20.2 GB

Ranked by distance from Recursive Flint 40B's parameter count. Full benchmark board.

All benchmarks Methodology

GPU	Model	Quant	Backend	Ctx	tok/s	prompt tok/s	VRAM GB	Source
RTX 5090 · 32 GB	Recursive Flint 40B	NVFP4	llama.cpp b11803	262144	224.8 ✓	—	—	src
RTX 5090 · 32 GB	Recursive Flint 40B	Q4_1	llama.cpp b11802 + Flint PR merged	262144	124.6 ✓	—	—	src