Muse-Spark-Open-816B-A42B

2026-10-15

Date the weights (or, if earlier, the technical report) were first made public, YYYY-MM-DD, as dated by the publisher. A Hugging Face repository creation date is used only when no dated publisher page exists, with a note.

other (Muse Model License)

License of the released weights as declared by the publisher (model card metadata or license file), e.g. apache-2.0, mit, or the name of a custom license.

text + image + video + audio

Input modalities of the released checkpoint as the publisher states them, from: text, image, video, audio, joined with ' + ' in that order ("text", "text + image + video").

816B parameters

Total parameter count as reported by the publisher. See params.note for whether embeddings, MTP heads or vision towers are included.

42B parameters

Parameters used per token in the forward pass, as reported by the publisher.

Not disclosed

Active parameters excluding the input embedding and the output head (unembedding). Stated by the publisher, or computed from params.active by subtracting vocab × d_model for each of the embedding and output head that params.active is stated to include (basis inferred, note with the arithmetic). Not disclosed when the scope of params.active is not stated.

Whether 816B includes the 1.8B-param Perception Encoder is not stated.

What the publisher's counts include or exclude (MTP modules, embeddings, vision encoder, n-gram tables).

3.31B parameters

Inferred from the source; see the explanation below.

Parameters in input embedding and output head, as reported.

62

Number of transformer blocks in the main backbone (attention or token mixer plus FFN), excluding MTP modules.

8,192

Residual stream width.

202,048

Embedding table rows (vocab_size in config, may include padding).

No

Input embedding and output head share weights.

132.129

Calculated: d_model / layers

Width-to-depth ratio, d_model divided by the number of blocks.

GQA, SWA

Token-mixing layer types in the backbone, e.g. GQA, MLA, SWA, Gated DeltaNet, Mamba-2, DSA (DeepSeek sparse attention).

1 SWA, then 1 global, then (3 SWA : 1 global) × 15

Inferred from the source; see the explanation below.

Token-mixer order, summarized as a pattern or layer positions. The fact's optional sequence lists each mixer in order; a ratio alone does not establish the order. Auxiliary prediction modules and standalone FFNs are excluded.

16

Inferred from the source; see the explanation below.

Layers whose KV cache grows with full context (dense or sparse global attention, MLA). Excludes sliding-window and linear/recurrent layers.

2.875

Calculated: attn.local_layers / attn.global_layers

attn.local_layers divided by attn.global_layers, the sliding-window to global interleave ratio (5 for '5 SWA : 1 global').

46

Inferred from the source; see the explanation below.

Layers with a fixed attention window.

128

Query heads in the global-attention layers.

16

Key/value heads in the global-attention layers. Not applicable to MLA, whose cache is a shared latent.

8,192

Calculated: heads × head_dim

attn.heads × attn.head_dim, the query projection width of global layers.

8

Calculated: heads / kv_heads

attn.heads divided by attn.kv_heads in global layers.

64

Query/key dimension per head in global-attention layers (for MLA, nope + rope dims).

2.048K tokens

Sliding-window size of the local layers.

16

KV heads in sliding-window layers (stated even when equal to attn.kv_heads).

128

Query heads in sliding-window layers (stated even when equal to attn.heads).

64

Query/key head dimension in sliding-window layers. Value head dimensions may differ; see attn.local_v_head_dim.

RoPE (SWA, θ 500000), NoPE (global)

Positional scheme in global layers: RoPE, partial RoPE, NoPE, or mixed per layer type.

500,000

RoPE base frequency of the sliding-window layers (stated even when equal to attn.rope_theta).

Yes

Queries and keys are normalized per head before the dot product.

elementwise sigmoid gate computed from the block input, applied after a weightless per-head output RMSNorm, before o_proj

Gate applied to the attention output before the output projection, with its placement.

1/sqrt(head_dim) × 5.4730064863875 (Q multiplied by qk_scale_factor after QK-norm)

Query-key scaling when it is not 1/sqrt(head_dim), e.g. query_pre_attn_scalar, length-dependent (SSMax-style) scaling, or value scaling.

No

Query/key/value projections have bias terms.

SwiGLU

FFN activation (SwiGLU, GeGLU, ...).

none

Clamping of GLU gate / up activations (e.g. SwiGLU limit 7), with the layers it applies to.

No

Inferred from the source; see the explanation below.

FFN projections have bias terms.

256

Routed experts per MoE layer.

8

Routed experts selected per token.

0

Inferred from the source; see the explanation below.

Always-on experts per MoE layer.

4,096

Intermediate width of one routed expert.

0

Number of initial blocks that use a dense FFN instead of MoE.

sigmoid

Gating function over expert logits (softmax, sigmoid).

Not disclosed

Load-balancing method, with coefficients where disclosed.

1

Multiplier applied to the combined routed-expert output (routed_scaling_factor).

Yes

Selected expert weights are renormalized to sum to 1.

4,096

Width of the projected space routed experts operate in (LatentMoE), when smaller than d_model.

2

Calculated: d_model / moe.latent_dim

shape.d_model divided by moe.latent_dim, how much LatentMoE compresses the hidden state before routed experts.

0.5

Calculated: moe.expert_dim / d_model

moe.expert_dim divided by shape.d_model.

32

Calculated: experts / top_k

Routed experts divided by experts selected per token.

0.0514706

Calculated: params.active / params.total

params.active divided by params.total.

Not disclosed

Additional future tokens predicted during pretraining (0 when the model was trained without MTP).

RMSNorm (1+w gain)

Normalization layer type in the blocks.

sandwich

Pre-norm, post-norm, or sandwich (pre and post).

20

tanh soft-capping value on output logits.

1e-05

Epsilon of the block normalization layers.

learned per-channel residual gate: alpha·residual + beta·h, beta = sigmoid(g/0.3), alpha = sqrt(sigmoid(-g/0.3)·(1+beta))

Residual design when it is not a plain sum, e.g. hyper-connections with n streams, gated residuals, depth-scaled residuals, FP32 residual accumulation.

weightless RMSNorm after embedding lookup

Scaling or normalization applied to input embeddings (e.g. × sqrt(d_model), RMSNorm after embedding).

logits × 1/sqrt(d_model/256) = 0.1767766952966369, before the softcap

Scaling applied to output logits other than a softcap (e.g. muP-style width multiplier).

Not disclosed

Sequence length of the main pretraining stage.

Not disclosed

Longest sequence length trained on after context extension.

1M tokens

Longest context the publisher says the model supports, possibly beyond trained length via RoPE scaling.

Not disclosed

Tokens seen in pretraining including annealing and context extension, as reported.

Not disclosed

Optimizer used for the bulk of parameters.

Not disclosed

Peak learning rate of the main pretraining stage.

Not disclosed

Learning rate at the end of the decay.

Not disclosed

Learning-rate schedule shape with phase boundaries.

Not disclosed

Warmup length in steps or tokens.

Not disclosed

Global batch size, including any batch-size ramp: sequences × sequence length (token total), or the token total alone when only that is stated.

Not disclosed

Decoupled weight decay coefficient.

Not disclosed

(beta1, beta2) of the Adam-family optimizer or its AdamW fallback.

Not disclosed

Global gradient-norm clipping threshold.

Not disclosed

Matmul precision of pretraining (BF16, FP8 recipe, NVFP4 recipe) with the scaling granularity where disclosed. Never inferred from the released checkpoint dtype.

text, images, audio and video; more than 100 languages

Pretraining data composition as disclosed (domain shares, synthetic fraction, repetition).

68,480 bytes

Calculated: kv.bytes_at_context / 131072 tokens; uniform BF16, payload only; quantization overhead excluded. Global KV: 16 × (2048 × 16/8) B/entry; SWA KV: 46 × (2048 × 16/8) B/entry

Retained KV cache divided by context length, including bounded SWA caches and indexer keys. Exports use uniform BF16 at 131,072 tokens. The website can select uniform payload precision (no quantization overhead) or sourced component formats (including scales and documented padding). This is cache payload, not peak prefill memory or persistent prefix storage.

65,536 bytes

Calculated: 65536 B / 1 tokens (Global KV); uniform BF16, payload only; quantization overhead excluded. Global KV: 16 × (2048 × 16/8) B/entry; SWA KV: 46 × (2048 × 16/8) B/entry

Long-context global KV bytes per token, including indexer keys; excludes bounded SWA caches. Compression rates and cross-layer cache reuse are included. Exports use uniform BF16; the website's storage selector also applies here. This is the field to compare with global-KV figures in reports, using the matching storage recipe.

1,630 bytes

Calculated: kv.bytes_per_token (131072 tokens) / (params.active / 1e9); uniform BF16, payload only; quantization overhead excluded. Global KV: 16 × (2048 × 16/8) B/entry; SWA KV: 46 × (2048 × 16/8) B/entry

kv.bytes_per_token divided by active parameters in billions (131,072 tokens in exports).

213,709,775 bytes

Calculated: kv.bytes_at_context (131072 tokens) / (params.active / 1e9); uniform BF16, payload only; quantization overhead excluded. Global KV: 16 × (2048 × 16/8) B/entry; SWA KV: 46 × (2048 × 16/8) B/entry

kv.bytes_at_context divided by active parameters in billions (131,072 tokens in exports).

8,975,810,560 bytes

Calculated: 65536 B × 131072 (Global KV) + 188416 B × min(131072, 2048) (SWA KV); uniform BF16, payload only; quantization overhead excluded. Global KV: 16 × (2048 × 16/8) B/entry; SWA KV: 46 × (2048 × 16/8) B/entry

Retained KV cache for one sequence at the selected context, including SWA windows and indexer keys. Exports use uniform BF16 at 131,072 tokens. Excludes recurrent state, compressor tail state, draft/MTP caches, temporary prefill tensors, allocator metadata and persistent prefix snapshots. Does not estimate peak prefill or total serving memory.

39

Artificial Analysis Intelligence Index v4.3.2 of the post-trained model, from a dated Wayback snapshot of its Artificial Analysis page. The note names the variant (reasoning setting).