diff --git a/vllm_ascend/spec_decode/eagle_proposer.py b/vllm_ascend/spec_decode/eagle_proposer.py
index 356efad6..7c4f2e3b 100644
--- a/vllm_ascend/spec_decode/eagle_proposer.py
+++ b/vllm_ascend/spec_decode/eagle_proposer.py
@@ -45,6 +45,9 @@ class EagleProposer(Proposer):
         self.vllm_config = vllm_config
         self.device = device
         self.runner = runner
+        self.speculative_config = vllm_config.speculative_config
+        self.draft_model_config = self.speculative_config.draft_model_config
+        self.method = self.speculative_config.method
 
         self.block_size = vllm_config.cache_config.block_size
         # We need to get the hidden size from the draft model config because
@@ -99,6 +102,29 @@ class EagleProposer(Proposer):
                                        device="cpu",
                                        dtype=torch.int32)
         self.attn_mask_builder = AttentionMaskBuilder(self.device)
+        self.eagle3_use_aux_hidden_state: bool = (
+            self._get_eagle3_use_aux_hidden_state_from_config())
+
+    def _get_eagle3_use_aux_hidden_state_from_config(self) -> bool:
+        """
+        NOTE(2025-12-18): This is an explicit copy from vLLM EagleProposer, only added
+        to align with its logics.
+
+        Some eagle3 heads (e.g., nvidia/gpt-oss-120b-Eagle3-v2) do not use auxiliary
+        hidden states and directly uses the last layer output just like eagle1.
+        They might indicate this by setting "use_aux_hidden_state" to False
+        inside the "eagle_config" dict of their hf_config.
+        """
+        if self.method != "eagle3":
+            return False
+        # Assume that eagle3 heads use aux hidden states by default
+        use_aux_hidden_state = True
+        eagle_config = getattr(self.draft_model_config.hf_config,
+                               "eagle_config", None)
+        if eagle_config is not None:
+            use_aux_hidden_state = eagle_config.get("use_aux_hidden_state",
+                                                    True)
+        return use_aux_hidden_state
 
     def load_model(self, model: nn.Module) -> None:
         target_attn_layer_names = set(
diff --git a/vllm_ascend/worker/model_runner_v1.py b/vllm_ascend/worker/model_runner_v1.py
index 86036a17..5e1cc0fc 100644
--- a/vllm_ascend/worker/model_runner_v1.py
+++ b/vllm_ascend/worker/model_runner_v1.py
@@ -112,7 +112,6 @@ from vllm_ascend.sample.logits_processor import build_logitsprocs
 from vllm_ascend.sample.sampler import AscendSampler
 from vllm_ascend.spec_decode import get_spec_decode_method
 from vllm_ascend.spec_decode.eagle_proposer import EagleProposer
-from vllm_ascend.spec_decode.interface import SpecDcodeType
 from vllm_ascend.spec_decode.mtp_proposer import MtpProposer
 from vllm_ascend.utils import (AscendDeviceType, ProfileExecuteDuration,
                                enable_sp, get_ascend_device_type, is_moe_model,
@@ -385,6 +384,10 @@ class NPUModelRunner(GPUModelRunner):
             )
             if get_pp_group().is_last_rank:
                 self.drafter = self._get_drafter()
+                if self.speculative_config.method == "eagle3":
+                    assert isinstance(self.drafter, EagleProposer)
+                    self.use_aux_hidden_state_outputs = (
+                        self.drafter.eagle3_use_aux_hidden_state)
                 self.rejection_sampler = RejectionSampler(self.sampler)
             self.actual_seq_lengths_q = list(
                 range(self.decode_token_per_req, self.max_num_tokens + 1,
@@ -1417,7 +1420,7 @@ class NPUModelRunner(GPUModelRunner):
                 scheduler_output)
 
             aux_hidden_states = None
-            if self.drafter and self.drafter.name == SpecDcodeType.EAGLE3:
+            if self.use_aux_hidden_state_outputs:
                 hidden_states, aux_hidden_states = hidden_states
 
         kv_connector_output = KVConnectorOutput(
@@ -1929,7 +1932,7 @@ class NPUModelRunner(GPUModelRunner):
                     update_attn_params(self.update_stream, forward_context,
                                        num_tokens, self.vllm_config)
 
-        if self.drafter and self.drafter.name == SpecDcodeType.EAGLE3:
+        if self.use_aux_hidden_state_outputs:
             hidden_states, _ = hidden_states
         else:
             hidden_states = hidden_states
@@ -2196,7 +2199,7 @@ class NPUModelRunner(GPUModelRunner):
             if self.drafter:
                 logger.info("Loading drafter model...")
                 self.drafter.load_model(self.model)
-                if self.drafter.name == SpecDcodeType.EAGLE3:
+                if self.use_aux_hidden_state_outputs:
                     self.model.set_aux_hidden_state_layers(
                         self.model.get_eagle3_aux_hidden_state_layers())