Step-5-Preview

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 (StepFun Model License 1.0)

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

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").

600B parameters

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

27B 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.

599.1B parameters

Calculated: embedding + output head + all experts + other weights, stepfun-ai/Step-5-Preview@364b3b0 safetensors headers; reproduces stated total (backbone) and active (with embedding)

Parameters of the text backbone counted from the released checkpoint's safetensors headers (embedding, output head, all experts and other weights; excluding MTP modules, vision/audio towers and n-gram lookup tables). Shown only when the counts reproduce the publisher's stated total and active parameters within 5% under some scope.

1.584B parameters

Calculated: input embedding + output head, stepfun-ai/Step-5-Preview@364b3b0 safetensors headers

Input embedding plus output head (once if tied), counted from the checkpoint.

597.5B parameters

Calculated: dense + routed backbone weights; excludes input embedding, output head, lookup tables, MTP and other modalities, stepfun-ai/Step-5-Preview@364b3b0 safetensors headers; reproduces stated total (backbone) and active (with embedding)

Language backbone weights excluding input embeddings, output head, lookup/Engram tables, MTP modules and vision/audio towers. Equals params.counted_total minus params.counted_embedding. Same validation as params.counted_total.

6.79B parameters

Calculated: max(sum of dense and routed weights sharing a full layers.<index> tensor path), all experts included, stepfun-ai/Step-5-Preview@364b3b0 safetensors headers

Largest group of language-backbone weights sharing a full layers.<index> tensor path, including every routed expert. Excludes lookup/Engram tables, embeddings, output head, MTP and other modalities. Hybrid checkpoints may store mixers and FFNs as separate layers. All-gather volume also depends on wrapping, sharding and dtype.

26.11B parameters

Calculated: non-expert weights + routed experts × 8/352, excluding embedding and output head, stepfun-ai/Step-5-Preview@364b3b0 safetensors headers; reproduces stated total (backbone) and active (with embedding)

Active parameters per token excluding the input embedding and output head, counted from the checkpoint as all non-expert weights plus routed-expert weights × top_k / experts. Same validation as params.counted_total.

Rounded publisher figures; inclusion of the 3 MTP layers and the vision encoder is not stated.

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

92

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

4,096

Residual stream width.

128,896

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

No

Input embedding and output head share weights.

44.5217

Calculated: d_model / layers

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

SWA, DSA (Sparse GQA)

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

3 SWA : 1 DSA

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.

23

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

3

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').

69

Layers with a fixed attention window.

64

Query heads in the global-attention layers.

4

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

12,288

Calculated: heads × head_dim

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

16

Calculated: heads / kv_heads

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

192

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

512 tokens

Sliding-window size of the local layers.

4

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

64

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

192

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

partial RoPE (DSA, 64 of 192 dims), RoPE (SWA)

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

10,000,000

RoPE base frequency of the global layers at release.

10,000

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

0.333333

Fraction of each head's QK dimensions that are rotated.

Yes

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

head-wise sigmoid on the attention output, before o_proj

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

DSA indexer (16 heads, proxy dim 256) scores 8-token region summaries; each query attends to its top-512 regions

Token-selection mechanism inside global layers, e.g. DSA lightning indexer with top-k.

No

Query/key/value projections have bias terms.

256

Width of the sparse-attention indexer key cached per token in each global layer (DSA lightning indexer), counted in the KV cache.

• DSA K/V

  Layers

  23

  Elements

  1536

  Tokens per entry

  1

  Source: code · inference/model.py; Step5Cache, DenseLayerCache

  retain full K/V and completed-region summaries

  4 KV heads × 192 dims, K and V.

• DSA indexer region summaries

  Layers

  23

  Elements

  256

  Tokens per entry

  8

  Source: code · inference/model.py; Step5Cache, DenseLayerCache

  retain full K/V and completed-region summaries

  proxy_dim 256 per 8-token region (region_block_size); stored FP8 in kernel.py DSALayerCache.

• SWA K/V

  Layers

  69

  Elements

  1536

  Tokens per entry

  1

  Window

  512

  Source: code · inference/model.py; Step5Cache, DenseLayerCache

  Ring buffer of the last ``window`` tokens for one sliding-attention layer.

  4 KV heads × 192 dims, K and V.

Cache owners and entry shapes for compressed or shared attention. Each item specifies layers owning the cache, scalar elements per entry (K and V together), tokens_per_entry (compression ratio, default 1), and optionally window (maximum entries retained). Completed compressed entries are floor(context / tokens_per_entry); incomplete compressor state is excluded. KV sizes use these groups when present.

SwiGLU

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

13,824

Intermediate width of the dense FFN layers that are actually executed.

SwiGLU limit 7 (silu(gate) ≤ 7, up in [−7, 7]) on all dense, shared and routed FFNs

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

No

FFN projections have bias terms.

3.375

Calculated: ffn.dim / d_model

ffn.dim divided by shape.d_model.

352

Routed experts per MoE layer.

8

Routed experts selected per token.

1

Always-on experts per MoE layer.

1,536

Intermediate width of one routed expert.

1,536

Total intermediate width of the shared expert(s).

3

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.

3

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

Yes

Selected expert weights are renormalized to sum to 1.

0.375

Calculated: moe.expert_dim / d_model

moe.expert_dim divided by shape.d_model.

44

Calculated: experts / top_k

Routed experts divided by experts selected per token.

0.045

Calculated: params.active / params.total

params.active divided by params.total.

0

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

Yes

The released checkpoint contains the MTP module weights.

RMSNorm (1+w gain)

Normalization layer type in the blocks.

pre

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

1e-05

Epsilon of the block normalization layers.

FP32 residual stream (hidden states and their gradients)

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

Not disclosed

Sequence length of the main pretraining stage.

Not disclosed

Longest sequence length trained on after context extension.

1.049M 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.

FP8 mixed precision

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

FP32 hidden states and gradients; FP32 RMSNorm gains; fine-grained gradient clamp (top 0.00004% of gradient values per layer)

Interventions used during pretraining to keep it stable (QK-clip, z-loss, logit caps, spike handling), with coefficients.

72,956 bytes

Calculated: kv.bytes_at_context / 131072 tokens; uniform BF16, payload only; quantization overhead excluded. DSA K/V: 23 × (1536 × 16/8) B/entry; DSA indexer region summaries: 23 × (256 × 16/8) B/entry; SWA K/V: 69 × (1536 × 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.

72,128 bytes

Calculated: 70656 B / 1 tokens (DSA K/V) + 11776 B / 8 tokens (DSA indexer region summaries); uniform BF16, payload only; quantization overhead excluded. DSA K/V: 23 × (1536 × 16/8) B/entry; DSA indexer region summaries: 23 × (256 × 16/8) B/entry; SWA K/V: 69 × (1536 × 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.

2,702 bytes

Calculated: kv.bytes_per_token (131072 tokens) / (params.active / 1e9); uniform BF16, payload only; quantization overhead excluded. DSA K/V: 23 × (1536 × 16/8) B/entry; DSA indexer region summaries: 23 × (256 × 16/8) B/entry; SWA K/V: 69 × (1536 × 16/8) B/entry

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

354,166,253 bytes

Calculated: kv.bytes_at_context (131072 tokens) / (params.active / 1e9); uniform BF16, payload only; quantization overhead excluded. DSA K/V: 23 × (1536 × 16/8) B/entry; DSA indexer region summaries: 23 × (256 × 16/8) B/entry; SWA K/V: 69 × (1536 × 16/8) B/entry

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

9,562,488,832 bytes

Calculated: 70656 B × 131072 (DSA K/V) + 11776 B × floor(131072 / 8) (DSA indexer region summaries) + 211968 B × min(131072, 512) (SWA K/V); uniform BF16, payload only; quantization overhead excluded. DSA K/V: 23 × (1536 × 16/8) B/entry; DSA indexer region summaries: 23 × (256 × 16/8) B/entry; SWA K/V: 69 × (1536 × 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.

44

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).