K2-Horizon-375B-A23B (mid_4_6000)

2026-09-03

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.

apache-2.0

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

Inferred from the source; see the explanation below.

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").

375B parameters

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

23B 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.

379.2B parameters

Calculated: embedding + output head + all experts + other weights, IFM/K2-Horizon-375B-A23B@7d22cc0 safetensors headers; reproduces stated total (backbone) and active (non-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.

3.08B parameters

Calculated: input embedding + output head, IFM/K2-Horizon-375B-A23B@7d22cc0 safetensors headers

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

376.1B parameters

Calculated: dense + routed backbone weights; excludes input embedding, output head, lookup tables, MTP and other modalities, IFM/K2-Horizon-375B-A23B@7d22cc0 safetensors headers; reproduces stated total (backbone) and active (non-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.

6.464B parameters

Calculated: max(sum of dense and routed weights sharing a full layers.<index> tensor path), all experts included, IFM/K2-Horizon-375B-A23B@7d22cc0 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.

23.59B parameters

Calculated: non-expert weights + routed experts × 8/192, excluding embedding and output head, IFM/K2-Horizon-375B-A23B@7d22cc0 safetensors headers; reproduces stated total (backbone) and active (non-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.

61

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

6,144

Residual stream width.

250,624

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

No

Input embedding and output head share weights.

100.721

Calculated: d_model / layers

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

GQA

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

61 GQA

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.

61

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

48

Query heads in the global-attention layers.

8

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

6,144

Calculated: heads × head_dim

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

6

Calculated: heads / kv_heads

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

128

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

partial RoPE (64 of 128 QK dims)

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.

0.5

Inferred from the source; see the explanation below.

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

No

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

No

Query/key/value projections have bias terms.

SwiGLU

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

16,384

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

No

FFN projections have bias terms.

2.66667

Calculated: ffn.dim / d_model

ffn.dim divided by shape.d_model.

192

Routed experts per MoE layer.

8

Routed experts selected per token.

1

Always-on experts per MoE layer.

1,792

Intermediate width of one routed expert.

1,792

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).

per-expert bias for top-k selection only (update rate 1e-3)

Load-balancing method, with coefficients where disclosed.

2.5

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

Yes

Selected expert weights are renormalized to sum to 1.

0.291667

Calculated: moe.expert_dim / d_model

moe.expert_dim divided by shape.d_model.

24

Calculated: experts / top_k

Routed experts divided by experts selected per token.

0.0613333

Calculated: params.active / params.total

params.active divided by params.total.

0

Inferred from the source; see the explanation below.

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

RMSNorm

Normalization layer type in the blocks.

pre

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

1e-06

Epsilon of the block normalization layers.

8.192K tokens

Sequence length of the main pretraining stage.

524.3K tokens

Longest sequence length trained on after context extension.

524.3K tokens

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

8K → 32K (1.09T) → 128K (503B) → 512K (117B, then 201B with agentic/reasoning SFT data)

How context was extended (stages, RoPE base change, YaRN factor).

17.04T tokens

Inferred from the source; see the explanation below.

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.

~33.5M tokens

Inferred from the source; see the explanation below.

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

33.5M tokens

Inferred from the source; see the explanation below.

Global batch size in tokens during the main pretraining stage, after any ramp (sequences × sequence length when stated in sequences, marked inferred).

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.

Phase 2 continues on newer and better data; Midtraining Stage 4 shifts the mix toward agentic and reasoning SFT data

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

• pretraining-phase-1

  Tokens

  7080000000000

  Seq len

  8192

  Steps

  211000

  Source: model-card · README.md; Training Overview table

  Pretraining — Phase 1 | 211000 | 7.08T | 8K | Pretraining.

• pretraining-phase-2

  Tokens

  8050000000000

  Seq len

  8192

  Steps

  240000

  Source: model-card · README.md; Training Overview table

  Pretraining — Phase 2 | 240000 | 8.05T | 8K | Continued pretraining from Phase 1, with newer and better data.

  Continued from Phase 1 with newer and better data.

• midtraining-1

  Tokens

  1090000000000

  Seq len

  32768

  Steps

  32500

  Source: model-card · README.md; Training Overview table

  Midtraining — Stage 1 | 32500 | 1.09T | 32K | Context extension.

  Context extension.

• midtraining-2

  Tokens

  503000000000

  Seq len

  131072

  Steps

  15000

  Source: model-card · README.md; Training Overview table

  Midtraining — Stage 2 | 15000 | 503B | 128K | Context extension.

  Context extension.

• midtraining-3

  Tokens

  117000000000

  Seq len

  524288

  Steps

  3500

  Source: model-card · README.md; Training Overview table

  Midtraining — Stage 3 | 3500 | 117B | 512K | Context extension.

  Context extension.

• midtraining-4

  Tokens

  201000000000

  Seq len

  524288

  Steps

  6000

  Source: model-card · README.md; Training Overview table

  Midtraining — Stage 4 | 6000 | 201B | 512K | Continued context extension from Stage 3, with the data mix shifted toward agentic and reasoning SFT data.

  Mix shifted toward agentic and reasoning SFT data; final checkpoint is tag mid_4_6000.

Ordered pretraining stages. Each item: name, tokens, seq_len, lr, batch_tokens, batch_sequences, steps, note, quote, and source when it differs from the fact's source.

45.4427

Calculated: train.tokens / params.total

train.tokens divided by params.total.

722.389

Calculated: train.tokens / params.counted_active_non_embedding

train.tokens divided by params.counted_active_non_embedding, or by the stated params.active_non_embedding when there is no validated checkpoint count.

249,856 bytes

Calculated: kv.bytes_at_context / 131072 tokens; uniform BF16, payload only; quantization overhead excluded. Global KV: 61 × (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.

249,856 bytes

Calculated: 249856 B / 1 tokens (Global KV); uniform BF16, payload only; quantization overhead excluded. Global KV: 61 × (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.

10,863 bytes

Calculated: kv.bytes_per_token (131072 tokens) / (params.active / 1e9); uniform BF16, payload only; quantization overhead excluded. Global KV: 61 × (2048 × 16/8) B/entry

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

1,423,875,027 bytes

Calculated: kv.bytes_at_context (131072 tokens) / (params.active / 1e9); uniform BF16, payload only; quantization overhead excluded. Global KV: 61 × (2048 × 16/8) B/entry

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

32,749,125,632 bytes

Calculated: 249856 B × 131072 (Global KV); uniform BF16, payload only; quantization overhead excluded. Global KV: 61 × (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.

31

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).