diff --git a/mlx_lm/models/cohere2_moe.py b/mlx_lm/models/cohere2_moe.py
new file mode 100644
index 000000000..3850c6247
--- /dev/null
+++ b/mlx_lm/models/cohere2_moe.py
@@ -0,0 +1,350 @@
+# Copyright © 2026 Apple Inc.
+
+from dataclasses import dataclass
+from typing import List, Optional, Tuple
+
+import mlx.core as mx
+import mlx.nn as nn
+
+from .base import BaseModelArgs, create_attention_mask, scaled_dot_product_attention
+from .cache import KVCache, RotatingKVCache
+from .switch_layers import SwitchGLU
+
+
+@dataclass
+class ModelArgs(BaseModelArgs):
+    model_type: str
+    hidden_size: int = 1024
+    head_dim: int = 128
+    num_hidden_layers: int = 36
+    intermediate_size: int = 1024
+    num_attention_heads: int = 64
+    num_key_value_heads: int = 8
+    rope_theta: float = 50000.0
+    vocab_size: int = 256000
+    layer_norm_eps: float = 1e-05
+    logit_scale: float = 0.0625
+    attention_bias: bool = False
+    layer_norm_bias: bool = False
+    sliding_window: int = 4096
+    sliding_window_pattern: int = 4
+    num_experts: int = 128
+    num_experts_per_tok: int = 8
+    norm_topk_prob: bool = True
+    num_shared_experts: Optional[int] = None
+    moe_num_shared_experts: int = 4
+    moe_gate_act: str = "sigmoid"
+    expert_selection_fn: Optional[str] = None
+    shared_expert_combination_strategy: str = "average"
+    rms_norm_eps: Optional[float] = None
+    first_k_dense_replace: int = 0
+    prefix_dense_intermediate_size: Optional[int] = None
+    prefix_dense_sliding_window_pattern: int = 1
+    layer_types: Optional[List[str]] = None
+
+    def __post_init__(self):
+        if self.num_shared_experts is not None:
+            self.moe_num_shared_experts = self.num_shared_experts
+        if self.expert_selection_fn is not None:
+            self.moe_gate_act = self.expert_selection_fn
+        if self.prefix_dense_intermediate_size is None:
+            self.prefix_dense_intermediate_size = self.intermediate_size
+
+
+def is_prefix_dense_layer(args: ModelArgs, layer_idx: int):
+    return layer_idx < args.first_k_dense_replace
+
+
+def is_sliding_layer(args: ModelArgs, layer_idx: int):
+    if is_prefix_dense_layer(args, layer_idx):
+        return False
+    if args.layer_types is not None:
+        return args.layer_types[layer_idx] == "sliding_attention"
+    return (layer_idx + 1) % args.sliding_window_pattern != 0
+
+
+def norm_layer(args: ModelArgs):
+    if args.rms_norm_eps is not None:
+        return nn.RMSNorm(args.hidden_size, eps=args.rms_norm_eps)
+    return nn.LayerNorm(
+        args.hidden_size, eps=args.layer_norm_eps, bias=args.layer_norm_bias
+    )
+
+
+class Attention(nn.Module):
+    def __init__(self, args: ModelArgs, layer_idx: int):
+        super().__init__()
+        self.args = args
+        self.layer_idx = layer_idx
+
+        dim = args.hidden_size
+        self.n_heads = n_heads = args.num_attention_heads
+        self.n_kv_heads = n_kv_heads = args.num_key_value_heads
+        self.head_dim = head_dim = args.head_dim
+        self.scale = head_dim**-0.5
+
+        attetion_bias = args.attention_bias
+
+        self.q_proj = nn.Linear(dim, n_heads * head_dim, bias=attetion_bias)
+        self.k_proj = nn.Linear(dim, n_kv_heads * head_dim, bias=attetion_bias)
+        self.v_proj = nn.Linear(dim, n_kv_heads * head_dim, bias=attetion_bias)
+        self.o_proj = nn.Linear(n_heads * head_dim, dim, bias=attetion_bias)
+
+        self.rope = nn.RoPE(head_dim, traditional=True, base=args.rope_theta)
+
+        self.use_sliding_window = is_sliding_layer(args, layer_idx)
+        self.force_rope = (
+            is_prefix_dense_layer(args, layer_idx)
+            and args.prefix_dense_sliding_window_pattern == 1
+        )
+
+    def __call__(
+        self,
+        x: mx.array,
+        mask: Optional[mx.array] = None,
+        cache: Optional[Tuple[mx.array, mx.array]] = None,
+    ) -> mx.array:
+        B, L, D = x.shape
+
+        queries, keys, values = self.q_proj(x), self.k_proj(x), self.v_proj(x)
+
+        queries = queries.reshape(B, L, self.n_heads, -1).transpose(0, 2, 1, 3)
+        keys = keys.reshape(B, L, self.n_kv_heads, -1).transpose(0, 2, 1, 3)
+        values = values.reshape(B, L, self.n_kv_heads, -1).transpose(0, 2, 1, 3)
+
+        # Cohere2Moe applies RoPE to sliding layers and optionally to prefix
+        # dense full-attention layers.
+        if self.use_sliding_window or self.force_rope:
+            if cache is None:
+                queries = self.rope(queries)
+                keys = self.rope(keys)
+            else:
+                queries = self.rope(queries, offset=cache.offset)
+                keys = self.rope(keys, offset=cache.offset)
+
+        if cache is not None:
+            keys, values = cache.update_and_fetch(keys, values)
+
+        sdpa_type = mx.float32 if queries.dtype == mx.float16 else queries.dtype
+        output = scaled_dot_product_attention(
+            queries.astype(sdpa_type),
+            keys,
+            values,
+            cache=cache,
+            scale=self.scale,
+            mask=mask,
+        ).astype(queries.dtype)
+        output = output.transpose(0, 2, 1, 3).reshape(B, L, -1)
+        return self.o_proj(output)
+
+
+class MLP(nn.Module):
+    def __init__(self, dim, hidden_dim):
+        super().__init__()
+        self.gate_proj = nn.Linear(dim, hidden_dim, bias=False)
+        self.up_proj = nn.Linear(dim, hidden_dim, bias=False)
+        self.down_proj = nn.Linear(hidden_dim, dim, bias=False)
+
+    def __call__(self, x):
+        return self.down_proj(nn.silu(self.gate_proj(x)) * self.up_proj(x))
+
+
+class CohereMoeSparseMoeBlock(nn.Module):
+    def __init__(self, args: ModelArgs):
+        super().__init__()
+        dim = args.hidden_size
+        intermediate_size = args.intermediate_size
+
+        self.num_experts = num_experts = args.num_experts
+        self.top_k = args.num_experts_per_tok
+        self.norm_topk_prob = args.norm_topk_prob
+
+        self.gate = nn.Linear(dim, num_experts, bias=False)
+        self.switch_mlp = SwitchGLU(dim, intermediate_size, num_experts)
+
+        if getattr(args, "moe_num_shared_experts", 0) > 0:
+            shared_intermediate_size = (
+                args.intermediate_size * args.moe_num_shared_experts
+            )
+            self.shared_experts = MLP(
+                args.hidden_size, shared_intermediate_size,
+            )
+            self.shared_expert_combination_strategy = \
+                args.shared_expert_combination_strategy
+            assert self.shared_expert_combination_strategy in [
+                "average", "sum"
+            ], "shared_expert_combination_strategy "
+            "must be one of ['average', 'sum']"
+        else:
+            self.shared_experts = None
+            self.shared_expert_combination_strategy = None
+
+        if args.moe_gate_act == "softmax":
+            self.gate_act = nn.Softmax()
+        elif args.moe_gate_act == "sigmoid":
+            self.gate_act = nn.Sigmoid()
+        else:
+            raise ValueError(f"{args.moe_gate_act} is not supported.")
+
+    def __call__(
+        self,
+        x: mx.array,
+    ):
+        gates = self.gate(x)
+        gates = self.gate_act(gates.astype(mx.float32))
+
+        k = self.top_k
+        inds = mx.stop_gradient(mx.argpartition(-gates, kth=k - 1, axis=-1)[..., :k])
+        scores = mx.take_along_axis(gates, inds, axis=-1)
+        if self.norm_topk_prob:
+            scores = scores / mx.maximum(scores.sum(axis=-1, keepdims=True), 1e-12)
+
+        y = self.switch_mlp(x, inds)
+        y = (y * scores[..., None]).sum(axis=-2).astype(y.dtype)
+
+        if self.shared_experts is not None:
+            if self.shared_expert_combination_strategy == "average":
+                y = (y + self.shared_experts(x)) / 2
+            else:
+                y = y + self.shared_experts(x)
+
+        return y
+
+
+class CohereMoEDecoderLayer(nn.Module):
+    def __init__(self, args: ModelArgs, layer_idx: int):
+        super().__init__()
+        self.hidden_size = args.hidden_size
+        self.n_heads = args.num_attention_heads
+
+        self.self_attn = Attention(args, layer_idx)
+        self.mlp = (
+            MLP(args.hidden_size, args.prefix_dense_intermediate_size)
+            if is_prefix_dense_layer(args, layer_idx)
+            else CohereMoeSparseMoeBlock(args)
+        )
+        self.input_layernorm = norm_layer(args)
+        self.args = args
+
+    def __call__(
+        self,
+        x: mx.array,
+        mask: Optional[mx.array] = None,
+        cache: Optional[Tuple[mx.array, mx.array]] = None,
+    ) -> mx.array:
+
+        h = self.input_layernorm(x)
+        attn_h = self.self_attn(h, mask, cache)
+        ff_h = self.mlp(h)
+
+        return attn_h + ff_h + x
+
+
+class CohereModel(nn.Module):
+    def __init__(self, args: ModelArgs):
+        super().__init__()
+        self.args = args
+        self.vocab_size = args.vocab_size
+        self.num_hidden_layers = args.num_hidden_layers
+        assert self.vocab_size > 0
+        self.window_size = args.sliding_window
+        self.embed_tokens = nn.Embedding(args.vocab_size, args.hidden_size)
+        self.layers = [
+            CohereMoEDecoderLayer(args=args, layer_idx=i)
+            for i in range(args.num_hidden_layers)
+        ]
+        self.norm = norm_layer(args)
+
+    def __call__(
+        self,
+        inputs: mx.array,
+        cache=None,
+    ):
+        h = self.embed_tokens(inputs)
+
+        if cache is None:
+            cache = [None] * len(self.layers)
+
+        for layer, c in zip(self.layers, cache):
+            mask = create_attention_mask(
+                h,
+                c,
+                window_size=(
+                    self.window_size if layer.self_attn.use_sliding_window else None
+                ),
+            )
+
+            h = layer(h, mask, c)
+
+        return self.norm(h)
+
+
+class Model(nn.Module):
+    def __init__(self, args: ModelArgs):
+        super().__init__()
+        self.model_type = args.model_type
+        self.model = CohereModel(args)
+        self.args = args
+
+    def __call__(
+        self,
+        inputs: mx.array,
+        cache=None,
+    ):
+        out = self.model(inputs, cache)
+        out = self.model.embed_tokens.as_linear(out)
+        out = out * self.model.args.logit_scale
+        return out
+
+    def make_cache(self):
+        caches = []
+        for i in range(self.args.num_hidden_layers):
+            if is_sliding_layer(self.args, i):
+                caches.append(
+                    RotatingKVCache(max_size=self.args.sliding_window, keep=0)
+                )
+            else:
+                caches.append(KVCache())
+        return caches
+
+    def sanitize(self, weights):
+        for l in range(self.args.num_hidden_layers):
+            if is_prefix_dense_layer(self.args, l):
+                continue
+            prefix = f"model.layers.{l}"
+            for n in ["up_proj", "down_proj", "gate_proj"]:
+                for k in ["weight", "scales", "biases"]:
+                    if f"{prefix}.mlp.experts.0.{n}.{k}" in weights:
+                        to_join = [
+                            weights.pop(f"{prefix}.mlp.experts.{e}.{n}.{k}")
+                            for e in range(self.args.num_experts)
+                        ]
+                        weights[f"{prefix}.mlp.switch_mlp.{n}.{k}"] = mx.stack(to_join)
+
+        for key in list(weights.keys()):
+            if "rotary_emb.inv_freq" in key:
+                weights.pop(key)
+            elif key.endswith(".bias"):
+                if ".mlp." in key:
+                    weights.pop(key)
+                elif ".self_attn." in key and not self.args.attention_bias:
+                    weights.pop(key)
+                elif "layernorm" in key.lower() and not self.args.layer_norm_bias:
+                    weights.pop(key)
+
+        return weights
+
+    @property
+    def quant_predicate(self):
+        def predicate(path, module):
+            if ".self_attn." in path:
+                return False
+            if ".mlp.gate" in path and "gate_proj" not in path:
+                return False
+            return True
+
+        return predicate
+
+    @property
+    def layers(self):
+        return self.model.layers
diff --git a/mlx_lm/utils.py b/mlx_lm/utils.py
index ef3d266b9..a3d03579f 100644
--- a/mlx_lm/utils.py
+++ b/mlx_lm/utils.py
@@ -172,6 +172,169 @@ def _transform_awq_weights(
     return new_weights, mlx_quantization
 
 
+def _e4m3_decode_table() -> mx.array:
+    """Return a 256-entry ``float32`` LUT mapping every E4M3FN byte to its value.
+
+    OCP ``E4M3FN``: 1 sign / 4 exponent (bias 7) / 3 mantissa bits, no infinities, and
+    ``(exp=15, mant=7)`` reserved for NaN (max finite magnitude is therefore 448). This
+    matches the byte convention MLX uses for ``nvfp4`` scales (verified empirically).
+    """
+    table = []
+    for b in range(256):
+        s = (b >> 7) & 1
+        e = (b >> 3) & 0xF
+        m = b & 0x7
+        if e == 0:
+            val = (m / 8.0) * 2.0**-6  # subnormal
+        elif e == 15 and m == 7:
+            val = float("nan")
+        else:
+            val = (1.0 + m / 8.0) * 2.0 ** (e - 7)
+        table.append(-val if s else val)
+    return mx.array(table, dtype=mx.float32)
+
+
+# Built once; reused by every NVFP4 fold.
+_E4M3_DECODE_LUT = _e4m3_decode_table()
+
+
+def _f32_to_e4m3(x: mx.array) -> mx.array:
+    """Encode non-negative ``float32`` values to ``E4M3FN`` bytes (round-to-nearest-even).
+
+    Pure-MLX bit manipulation (MLX exposes no float8 dtype). Saturates to 448 on
+    overflow and flushes to the subnormal grid / zero on underflow. Inputs are assumed
+    ``>= 0`` (NVFP4 group scales are magnitudes), so the sign bit is always 0.
+    """
+    x = mx.maximum(x.astype(mx.float32), 0.0)
+    bits = x.view(mx.uint32)
+    fexp = (bits >> 23) & 0xFF  # fp32 exponent, bias 127
+    fman = bits & 0x7FFFFF  # fp32 mantissa, 23 bits
+
+    # Normal path: target E4M3 biased exponent e = (fexp - 127) + 7.
+    e = fexp.astype(mx.int32) - 120
+    drop = 20  # 23 -> 3 mantissa bits
+    round_bit = (fman >> (drop - 1)) & 1
+    sticky = (fman & ((1 << (drop - 1)) - 1)) != 0
+    mant3 = fman >> drop
+    roundup = round_bit & (sticky.astype(mx.uint32) | (mant3 & 1))
+    mant3 = mant3 + roundup
+    carry = mant3 >> 3  # mantissa overflowed past 7 -> bump exponent
+    mant3 = mant3 & 0x7
+    e = e + carry.astype(mx.int32)
+    # Saturate: e > 15, or the NaN slot (e == 15, mant == 7), clamps to 448 (e15 m6).
+    over = (e > 15) | ((e == 15) & (mant3 == 7))
+    e = mx.where(over, mx.array(15, mx.int32), e)
+    mant3 = mx.where(over, mx.array(6, mx.uint32), mant3)
+    normal_byte = (e.astype(mx.uint32) << 3) | mant3
+    normal_valid = e >= 1
+
+    # Subnormal path: value = m * 2^-9, so m = round(x * 512) (RNE). m == 8 lands
+    # exactly on the smallest normal (0x08 = e1 m0 = 2^-6), so it is encoded correctly.
+    sub = x * 512.0
+    sub_floor = mx.floor(sub)
+    frac = sub - sub_floor
+    sf = sub_floor.astype(mx.uint32)
+    up = (frac > 0.5) | ((frac == 0.5) & ((sf & 1) == 1))
+    sub_byte = sf + up.astype(mx.uint32)
+
+    byte = mx.where(normal_valid, normal_byte, sub_byte)
+    return byte.astype(mx.uint8)
+
+
+def _transform_compressed_tensors_nvfp4_weights(
+    weights: Dict[str, mx.array],
+    quantization_config: Dict[str, Any],
+) -> Dict[str, mx.array]:
+    """Fold compressed-tensors NVFP4 weights into MLX-native ``nvfp4`` weights.
+
+    A ``nvfp4-pack-quantized`` checkpoint stores, per quantized Linear:
+
+    - ``<p>.weight_packed``       ``uint8``  ``[out, in // 2]``  (2x E2M1 per byte)
+    - ``<p>.weight_scale``        ``uint8``  ``[out, in // 16]`` (E4M3 per group of 16,
+      loaded by ``mx.load`` as raw bytes -- the same byte layout MLX uses for nvfp4 scales)
+    - ``<p>.weight_global_scale`` ``float32`` ``[1]`` (per-tensor; the real weight is
+      ``fp4 * weight_scale / weight_global_scale``)
+
+    MLX ``nvfp4`` ``QuantizedLinear`` expects ``<p>.weight`` (``uint32``) plus
+    ``<p>.scales`` (``uint8`` E4M3) and is single-level: the per-tensor global scale is
+    not representable (and is rejected on the Metal backend). Both decodes are linear in
+    the FP4 codes, so the global scale can be folded directly into the per-group E4M3
+    scales: ``scale_mlx = E4M3(weight_scale / global_scale)``. We keep the original packed
+    E2M1 codes bit-exact (only the per-group scales are re-encoded once), avoiding the
+    weight dequantize/re-quantize round-trip entirely.
+    """
+    packed_suffix = ".weight_packed"
+
+    new_weights = {}
+    for key in list(weights.keys()):
+        if key.endswith(packed_suffix):
+            prefix = key[: -len(packed_suffix)]
+            packed = weights[key]
+            scale = weights[f"{prefix}.weight_scale"]
+            global_scale = weights[f"{prefix}.weight_global_scale"].astype(mx.float32)
+
+            # weight_packed is uint8 [out, in//2]; reinterpret as uint32 [out, in//8]
+            # to match MLX's nvfp4 weight layout (bit-identical, no data movement).
+            new_weights[f"{prefix}.weight"] = packed.view(mx.uint32)
+
+            # Fold the per-tensor global scale into the per-group E4M3 scales:
+            # decode E4M3 -> divide by global scale -> re-encode E4M3. The FP4 codes are
+            # untouched, so this only re-rounds the (much smaller) scale tensor once.
+            decoded = _E4M3_DECODE_LUT[scale.astype(mx.uint32)]
+            new_weights[f"{prefix}.scales"] = _f32_to_e4m3(decoded / global_scale)
+        elif key.endswith(".weight_scale") or key.endswith(".weight_global_scale"):
+            # Consumed alongside their ``.weight_packed``.
+            continue
+        else:
+            new_weights[key] = weights[key]
+
+    return new_weights
+
+
+def _transform_compressed_tensors_int4_weights(
+    weights: Dict[str, mx.array],
+    quantization_config: Dict[str, Any],
+) -> Dict[str, mx.array]:
+    """Remap compressed-tensors INT4 ``pack-quantized`` weights to MLX affine weights.
+
+    A symmetric int4 ``pack-quantized`` checkpoint stores, per quantized Linear:
+
+    - ``<p>.weight_packed`` ``int32`` ``[out, in // 8]`` (8x int4 per word, LSB-first)
+    - ``<p>.weight_scale``  (``bf16``/``float``) ``[out, in // group_size]``
+    - ``<p>.weight_shape``  ``int64`` ``[2]`` (unused by MLX)
+
+    MLX affine ``QuantizedLinear`` uses the same int4 packing and dequantizes as
+    ``w * scale + bias``. Symmetric int4 stores values in ``[0, 15]`` representing
+    ``[-8, 7]``, i.e. ``value = packed - 8``, so we set ``bias = -8 * scale``. The packed
+    ``int32`` is bit-identical to MLX's ``uint32`` layout (reinterpreted via ``view``).
+    This is the model-agnostic version of the int4 remap in ``deepseek_v3.sanitize``.
+    """
+    weights_cfg = (
+        quantization_config.get("config_groups", {})
+        .get("group_0", {})
+        .get("weights", {})
+    )
+    group_size = weights_cfg.get("group_size", 32)
+    bits = weights_cfg.get("num_bits", 4)
+    packed_suffix = ".weight_packed"
+
+    new_weights = {}
+    for key in list(weights.keys()):
+        if key.endswith(packed_suffix):
+            prefix = key[: -len(packed_suffix)]
+            scale = weights[f"{prefix}.weight_scale"]
+            new_weights[f"{prefix}.weight"] = weights[key].view(mx.uint32)
+            new_weights[f"{prefix}.scales"] = scale
+            new_weights[f"{prefix}.biases"] = -(2 ** (bits - 1)) * scale
+        elif key.endswith(".weight_scale") or key.endswith(".weight_shape"):
+            # Consumed alongside their ``.weight_packed`` (shape is unused by MLX).
+            continue
+        else:
+            new_weights[key] = weights[key]
+
+    return new_weights, {"group_size": group_size, "bits": bits, "mode": "affine"}
+
+
 def _get_classes(config: dict):
     """
     Retrieve the model and model args classes based on the configuration.
@@ -342,6 +505,31 @@ def load_model(
 
     model = model_class(model_args)
 
+    # Transform compressed-tensors weights into MLX-native form before sanitize() so that
+    # per-expert (MoE) tensors are renamed to .weight/.scales(/.biases) and then stacked
+    # correctly. Skip if a model already remapped them itself (no .weight_packed left).
+    if (
+        (qc := config.get("quantization_config"))
+        and isinstance(qc, dict)
+        and qc.get("quant_method") == "compressed-tensors"
+        and any(k.endswith(".weight_packed") for k in weights)
+    ):
+        ct_format = qc.get("format")
+        if ct_format == "nvfp4-pack-quantized":
+            weights = _transform_compressed_tensors_nvfp4_weights(weights, qc)
+            quantization = {"group_size": 16, "bits": 4, "mode": "nvfp4"}
+        elif ct_format == "pack-quantized":
+            weights, quantization = _transform_compressed_tensors_int4_weights(
+                weights, qc
+            )
+        else:
+            raise ValueError(
+                f"Unsupported compressed-tensors format: {ct_format!r}. "
+                "Supported: 'nvfp4-pack-quantized', 'pack-quantized'."
+            )
+        config["quantization"] = quantization
+        config["quantization_config"] = quantization
+
     if hasattr(model, "sanitize"):
         weights = model.sanitize(weights)
 
diff --git a/tests/test_models.py b/tests/test_models.py
index 6e1fcd96e..75d74311b 100644
--- a/tests/test_models.py
+++ b/tests/test_models.py
@@ -1129,6 +1129,32 @@ def test_cohere(self):
             model, args.model_type, args.vocab_size, args.num_hidden_layers
         )
 
+    def test_cohere2_moe(self):
+        from mlx_lm.models import cohere2_moe
+
+        args = cohere2_moe.ModelArgs(
+            model_type="cohere2_moe",
+            hidden_size=64,
+            head_dim=16,
+            num_hidden_layers=4,
+            intermediate_size=128,
+            num_attention_heads=4,
+            num_key_value_heads=2,
+            vocab_size=1000,
+            sliding_window=4,
+            sliding_window_pattern=3,
+            num_experts=4,
+            num_experts_per_tok=2,
+            moe_num_shared_experts=0,
+            first_k_dense_replace=1,
+            prefix_dense_intermediate_size=96,
+            rms_norm_eps=1e-5,
+        )
+        model = cohere2_moe.Model(args)
+        self.model_test_runner(
+            model, args.model_type, args.vocab_size, args.num_hidden_layers
+        )
+
     def test_dbrx(self):
         from mlx_lm.models import dbrx