Gemma-4-31B

2026-04-02

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 + image

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").

30.7B parameters

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

30.7B parameters

Inferred from the source; see the explanation below.

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

29.29B parameters

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.

30.7B parameters

Calculated: embedding + output head + all experts + other weights, google/gemma-4-31B@5bbc2fb 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.

1.409B parameters

Calculated: input embedding + output head, google/gemma-4-31B@5bbc2fb safetensors headers

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

29.29B parameters

Calculated: dense + routed backbone weights; excludes input embedding, output head, lookup tables, MTP and other modalities, google/gemma-4-31B@5bbc2fb 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.

534M parameters

Calculated: max(sum of dense and routed weights sharing a full layers.<index> tensor path), all experts included, google/gemma-4-31B@5bbc2fb 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.

29.29B parameters

Calculated: all weights except embedding and output head, google/gemma-4-31B@5bbc2fb 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.

Language model only (1,410M embedder + 29,290M einsums). Excludes the 550M frozen vision encoder and the 500M MTP drafter.

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

1.41B parameters

Parameters in input embedding and output head, as reported.

29.29B parameters

Parameters excluding embedding and output head, as reported.

60

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

5,376

Residual stream width.

262,144

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

Yes

Input embedding and output head share weights.

SentencePiece, same as Gemini: split digits, preserved whitespace, byte-level encodings

Tokenizer type and lineage as described by the publisher (e.g. byte-level BPE shared with an earlier model).

89.6

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).

5 SWA : 1 global

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.

10

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

5

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').

50

Layers with a fixed attention window.

32

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.

16,384

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.

512

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

1.024K tokens

Sliding-window size of the local layers.

16

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

32

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

256

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

partial RoPE (global, p-RoPE p=0.25), RoPE (SWA)

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

1,000,000

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).

0.25

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

Yes

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

1.0 (no 1/sqrt(head_dim)); V RMS-normalized without learned scale

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.

GeGLU (tanh-approximate GELU)

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

21,504

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

No

FFN projections have bias terms.

4

Calculated: ffn.dim / d_model

ffn.dim divided by shape.d_model.

Not disclosed

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

RMSNorm

Normalization layer type in the blocks.

sandwich

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

30

tanh soft-capping value on output logits.

1e-06

Epsilon of the block normalization layers.

× sqrt(d_model)

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

Not disclosed

Sequence length of the main pretraining stage.

Not disclosed

Longest sequence length trained on after context extension.

262.1K 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.

Web documents (140+ languages), code, mathematics, images

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

January 2025

Pretraining data cutoff date.

10,240 TPUv6e chips (16 data × 16 sequence × 40 replica)

Accelerators used for pretraining, count and duration where stated.

10,240 TPU v6e

Accelerator count and type for the main pretraining run, as the count with thousands separators and the chip's short name ("2,048 H800", "2,048 GH200"). A stated range or bound stays as stated ("2,048–4,096 GH200").

data16 · seq16 · replica40

Parallelism layout of the main pretraining run as stated, as degree-annotated terms in the order TP, PP, VPP, EP, CP, DP ("PP16 · EP64 · ZeRO-1 DP"); a term without a stated degree has none ("FSDP").

88,320 bytes

Calculated: kv.bytes_at_context / 131072 tokens; uniform BF16, payload only; quantization overhead excluded. Global KV: 10 × (4096 × 16/8) B/entry; SWA KV: 50 × (8192 × 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.

81,920 bytes

Calculated: 81920 B / 1 tokens (Global KV); uniform BF16, payload only; quantization overhead excluded. Global KV: 10 × (4096 × 16/8) B/entry; SWA KV: 50 × (8192 × 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.

2,877 bytes

Calculated: kv.bytes_per_token (131072 tokens) / (params.active / 1e9); uniform BF16, payload only; quantization overhead excluded. Global KV: 10 × (4096 × 16/8) B/entry; SWA KV: 50 × (8192 × 16/8) B/entry

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

377,077,493 bytes

Calculated: kv.bytes_at_context (131072 tokens) / (params.active / 1e9); uniform BF16, payload only; quantization overhead excluded. Global KV: 10 × (4096 × 16/8) B/entry; SWA KV: 50 × (8192 × 16/8) B/entry

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

11,576,279,040 bytes

Calculated: 81920 B × 131072 (Global KV) + 819200 B × min(131072, 1024) (SWA KV); uniform BF16, payload only; quantization overhead excluded. Global KV: 10 × (4096 × 16/8) B/entry; SWA KV: 50 × (8192 × 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.

15

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).