Are Latent Reasoning Models Easily Interpretable?

Connor Dilgren; Sarah Wiegreffe

cs.LGApr 6, 2026

Are Latent Reasoning Models Easily Interpretable?

Authors: Connor Dilgren, Sarah Wiegreffe

Organizations: Department of Computer Science, University of Maryland, College Park, MD, USA

Abstract

Latent reasoning models (LRMs) have attracted significant research interest due to their low inference cost (relative to explicit reasoning models) and theoretical ability to explore multiple reasoning paths in parallel. However, these benefits come at the cost of reduced interpretability: LRMs are difficult to monitor because they do not reason in natural language. This paper presents an investigation into LRM interpretability by examining two state-of-the-art LRMs. First, we find that latent reasoning tokens are often unnecessary for LRMs' predictions; on logical reasoning datasets, LRMs can almost always produce the same final answers without using latent reasoning at all. This underutilization of reasoning tokens may partially explain why LRMs do not consistently outperform explicit reasoning methods and raises doubts about the stated role of these tokens in prior work. Second, we demonstrate that when latent reasoning tokens are necessary for performance, we can decode gold reasoning traces up to 65-93% of the time for correctly predicted instances. This suggests LRMs often implement the expected solution rather than an uninterpretable reasoning process. Finally, we present a method to decode a verified natural language reasoning trace from latent tokens without knowing a gold reasoning trace a priori, demonstrating that it is possible to find a verified trace for a majority of correct predictions but only a minority of incorrect predictions. Our findings highlight that current LRMs largely encode interpretable processes, and interpretability itself can be a signal of prediction correctness.

Figures & tables

Figure 1: An overview of our findings. Left: LRMs tend to commit to a final answer before exhausting their budget, indicating that they don’t effectively use all available reasoning tokens. Middle: Vocabulary projections of latent tokens often encode gold reasoning traces, suggesting that the model follows an interpretable reasoning trace rather than an opaque one. Right: We can generate candidate steps encoded by a latent token, and verify them by checking whether vocabulary projections changes as expected under modified prompts.

		GSM8k-Aug		PrOntoQA		ProsQA
Method	Base Model	Acc. (%)	# Tok.	Acc. (%)	# Tok.	Acc. (%)	# Tok.
No-CoT	GPT-2 Small	16.8 (16.5 † )	3.2	87.9 (93.8 † )	3.0	76.0 (76.7 † )	9.5
CoT	GPT-2 Small	41.6 (42.9 † )	31.0	99.3 (98.8 † )	92.7	74.2 (77.5 † )	51.6
Coconut	GPT-2 Small	33.1 (34.1 † )	9.2	99.0 (99.8 † )	9.0	98.0 (97.0 † )	15.5
CODI	GPT-2 Small	42.2 ∗ (43.7 ‡ )	12.3	95.1	12.0	81.6	18.2
No-CoT	Llama-3.2-1B	30.1 (30.9 ‡ )	4.2	99.8	3.0	87.8	8.6

Table 1: Model performance. Results from Hao et al. (2025) † and Shen et al. (2025b) ‡ shown in parentheses where available. See Appendix G for a discussion on the Coconut + Llama-3.2-1B-Instruct performance on GSM8k-Aug compared to the published result.

Figure 2: Early stopping results. Solid bars indicate the first match percentage, while hatched bars show the additional reasoning required for a stable match (black lines for one standard deviation), compared to the model’s full reasoning trace (RT).

LRM	Base Model	Reasoning Mode	GSM8k-Aug	PrOntoQA	ProsQA
Coconut	GPT-2 Small	No-CoT	0.3	0.0	0.0
	GPT-2 Small	CoT	19.2	0.0	-6.0
	Llama-3.2-1B-Instruct	No-CoT	-0.9	0.0	0.0
	Llama-3.2-1B-Instruct	CoT	29.3	-0.3	-1.4
CODI	GPT-2 Small	No-CoT	-5.9	-0.3	-0.2
	GPT-2 Small	CoT	3.5	-0.4	-7.4

Table 2: Relative performance of non-reasoning and explicit reasoning versus latent reasoning for the multi-reasoning models. Positive values mean latent reasoning underperformed.

Figure 3: Found gold reasoning trace in Coconut + GPT-2 Small’s vocabulary projections, from instance 220 of GSM8k-Aug’s test split. The model answered this question correctly.

Figure 4: Backtracking results. “Any Gold RT” includes additional solutions from the MultiChain GSM8k-Aug dataset ( Deng et al., 2025 ) . Darker bars exclude question tokens as operands, and lighter bars show the increase from including them as candidate operands.

Figure 5: Forward chaining results.

Appendix figures & tables26 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 6: Overview of the latent reasoning models Coconut and CODI. CODI will additionally pass each final-layer hidden state through a trained two-layer multi-layer perceptron before using it as the next input token during latent reasoning.

Dataset	Training	Validation	Original Test	Valid Gold Reasoning Trace Set	VP-Friendly Gold Reasoning Trace Set
GSM8k-Aug	385620	500	1319	1194	460
PrOntoQA	9000	200	800	-	-
ProsQA	17886	300	500	-	-

Appendix

Table 3: Dataset statistics. Model performance is calculated on the original test set. The early stopping experiment and backtracking experiment use the valid gold reasoning trace set, which filters out instances where the result of the last reasoning step is not equal to the correct answer. The forward chaining experiment uses the VP (Vocabulary Projection) friendly gold reasoning trace set, which filters for instances that use single-token numbers with unique operands and intermediate results in the gold reasoning trace.

Panel of Figure 7 — Figure 7: Example instances from each dataset.

	GPT-2 Small			Llama-3.2-1B-Instruct
Hyperparameter	GSM8k-Aug	ProsQA	PrOntoQA	GSM8k-Aug	ProsQA	PrOntoQA
Latent Tokens Per Stage	2	1	1	1	1	1
Stage 0 Epochs	3	5	5	3	3	3
Epochs Per Stage	3	5	5	1	1	2
Max Latent Stage	3	6	6	6	6	6
Total Epochs	50	50	50	10	10	20

Appendix

Table 4: Training hyperparameters for Coconut models. For the Coconut + GPT-2 Small model on GSM8k-Aug, the stage 0 training is replaced with checkpoint 6 of the CoT model.

	GPT-2 Small			Llama-3.2-1B-Instruct
Hyperparameter	GSM8k-Aug	ProsQA	PrOntoQA	GSM8k-Aug	ProsQA	PrOntoQA
Latent Loss Weight ( $\alpha$ )	1	1	1	1	1	1
CoT Loss Weight ( $\beta$ )	1	1	1	1	1	1
Distillation Loss Weight ( $\delta$ )	1	1	1	20	20	20
Num Latent Tokens	6	6	6	6	6	6
Total Epochs	40	40	40	10	10	10

Appendix

Table 5: Training hyperparameters for CODI models. For the CODI + GPT-2 Small model on GSM8k-Aug, we used a checkpoint released by the Shen et al. (2025b) authors.

	GPT-2 Small			Llama-3.2-1B-Instruct
Hyperparameter	GSM8k-Aug	ProsQA	PrOntoQA	GSM8k-Aug	ProsQA	PrOntoQA
Total Epochs	25	25	25	10	10	10
Batch Size	128	128	128	128	128	128
Learning Rate	$1\times 10^{-4}$	$1\times 10^{-4}$	$1\times 10^{-4}$	$5\times 10^{-5}$	$5\times 10^{-5}$	$5\times 10^{-5}$
Weight Decay	0.01	0.01	0.01	0.1	0.1	0.1
BF16 Precision	✗	✗	✗	✓	✓	✓

Appendix

Table 6: Training hyperparameters for CoT models.

	GPT-2 Small			Llama-3.2-1B-Instruct
Hyperparameter	GSM8k-Aug	ProsQA	PrOntoQA	GSM8k-Aug	ProsQA	PrOntoQA
Total Epochs	25	25	25	10	10	20
Batch Size	128	128	128	128	128	128
Learning Rate	$1\times 10^{-4}$	$1\times 10^{-4}$	$1\times 10^{-4}$	$5\times 10^{-5}$	$5\times 10^{-5}$	$5\times 10^{-5}$
Weight Decay	0.01	0.01	0.01	0.1	0.1	0.1
BF16 Precision	✗	✗	✗	✓	✓	✓

Appendix

Table 7: Training hyperparameters for No-CoT models.

	GPT-2 Small			Llama-3.2-1B-Instruct
Hyperparameter	GSM8k-Aug	ProsQA	PrOntoQA	GSM8k-Aug	ProsQA	PrOntoQA
Latent Loss Weight ( $\alpha$ )	1	1	1	1	1	1
CoT Loss Weight ( $\beta$ )	1	1	1	1	1	1
No-CoT Loss Weight ( $\gamma$ )	1	1	1	1	1	1
Latent Tokens Per Stage	2	1	1	1	1	1
Stage 0 Epochs	3	5	5	3	3	3

Appendix

Table 8: Training hyperparameters for multi-reasoning mode models which use Coconut for latent reasoning.

	GPT-2 Small			Llama-3.2-1B-Instruct
Hyperparameter	GSM8k-Aug	ProsQA	PrOntoQA	GSM8k-Aug	ProsQA	PrOntoQA
Latent Loss Weight ( $\alpha$ )	1	1	1	1	1	1
CoT Loss Weight ( $\beta$ )	1	1	1	1	1	1
No-CoT Loss Weight ( $\gamma$ )	1	1	1	1	1	1
Distillation Loss Weight ( $\delta$ )	1	1	1	20	20	20
Num Latent Tokens	6	6	6	6	6	6

Appendix

Table 9: Training hyperparameters for multi-reasoning mode models which use CODI for latent reasoning.

Base Model	Mode	Control Tokens
GPT-2 Small	No-CoT	[prompt] + [eocot] $\rightarrow$ answer
	CoT	[prompt] + [bocot] $\rightarrow$ cot + eocot $\rightarrow$ answer
	Latent	[prompt] + [bocot] $\rightarrow$ latent $\rightarrow$ [eocot] $\rightarrow$ answer
Llama-3.2-1B	No-CoT	[prompt] + [eot] + [eocot] $\rightarrow$ answer
	CoT	[prompt] + [eot] + [bocot] $\rightarrow$ cot + eocot $\rightarrow$ answer
	Latent	[prompt] + [eot] + [bocot] $\rightarrow$ latent $\rightarrow$ [eocot] $\rightarrow$ answer

Appendix

Table 10: Control tokens used to determine which reasoning mode the model uses. Tokens in square brackets are model inputs; they are not generated by the model. In CoT mode, the “eocot” is output by the model, while in latent mode, this token is an input. Llama-3.2-1B-Instruct uses the additional “end-of-turn” token after the prompt.

LRM	Base Model	Reasoning Mode	GSM8k-Aug	PrOntoQA	ProsQA
Coconut	GPT-2 Small	No-CoT	22.4	100.0	94.2
		CoT	41.3	100.0	88.2
		Coconut	22.1	100.0	94.2
	Llama-3.2-1B-Instruct	No-CoT	29.2	94.9	99.8
		CoT	59.4	94.6	98.4
		Coconut	30.1	94.9	99.8

Appendix

Table 11: Multi-reasoning model accuracy for each reasoning mode. This is the same data as in Table 2 , but with absolute performance instead of relative. Results from Cywiński et al. (2025) shown in parentheses where available.

		First Match (%)			Stable Match (%)
Base Model	Dataset	Explicit	Coconut	CODI	Explicit	Coconut	CODI
GPT-2 Small	GSM8k-Aug	98.1 $\pm$ 9.8	54.4 $\pm$ 36.2	44.2 $\pm$ 33.6	99.3 $\pm$ 5.4	69.0 $\pm$ 32.1	53.6 $\pm$ 32.5
	ProsQA	92.0 $\pm$ 21.7	0.0 $\pm$ 0.0	2.5 $\pm$ 12.6	97.7 $\pm$ 10.9	0.0 $\pm$ 0.0	4.0 $\pm$ 16.6
	PrOntoQA	45.5 $\pm$ 32.8	0.1 $\pm$ 2.9	0.3 $\pm$ 4.4	47.1 $\pm$ 33.0	0.1 $\pm$ 2.9	0.5 $\pm$ 5.8
Llama-3.2-1B	GSM8k-Aug	89.3 $\pm$ 19.1	19.5 $\pm$ 30.0	30.9 $\pm$ 33.9	90.6 $\pm$ 17.4	22.5 $\pm$ 32.5	37.7 $\pm$ 35.9
	ProsQA	71.1 $\pm$ 22.6	0.0 $\pm$ 0.7	0.1 $\pm$ 2.2	74.7 $\pm$ 22.5	0.0 $\pm$ 0.7	0.7 $\pm$ 7.4

Appendix

Table 12: Early stopping experiment results. This is the same data as Figure 2 .

Method	Base Model	GSM8k-Aug	PrOntoQA	ProsQA
		$\Delta$ Acc. (%)	$\Delta$ Acc. (%)	$\Delta$ Acc. (%)
Coconut	GPT-2 Small	$-26.0$	$0.0$	$0.0$
CODI	GPT-2 Small	$-31.5$	$0.2$	$0.2$
Coconut	Llama-3.2-1B	$-5.8$	$0.0$	$0.0$
CODI	Llama-3.2-1B	$-25.0$	$-0.1$	$0.0$

Appendix

Table 13: Change in accuracy when using mean latent tokens. Negative values indicate performance is lost when using mean latent tokens.

Figure 8: Probability and token rank of the model’s final answer during latent reasoning.

Figure 9: Rank-Biased Overlap (p=0.90), Top-k Jaccard similarity (k=10), and Jensen-Shannon Divergence between the model’s forced early answer distribution and its final answer distribution.

Figure 10: Found gold reasoning trace in CODI + GPT-2 Small’s vocabulary projections, from instance 36 of GSM8k-Aug’s test split. The CODI model does not encode numbers from the question in the latent tokens, at least not in a way that is detectable using vocabulary projection. The model answered this question correctly.

Figure 11: Found gold reasoning trace in Coconut + GPT-2 Small’s vocabulary projections, from instance 69 of GSM8k-Aug’s test split. The Coconut model encodes the correct final step, but it encodes an incorrect final step more strongly. The model seems to think that Bailey was losing

5 per week, rather than receiving

5 per week.

Figure 12: Coconut + GPT-2 Small’s vocabulary projections, from instance 179 of GSM8k-Aug’s test split. The Coconut model encodes a valid reasoning trace not contained in the set of known gold reasoning traces.

Figure 13: Coconut + GPT-2 Small’s vocabulary projections, from instance 229 of GSM8k-Aug’s test split. The model seems to encode the percentage 30% as simply 30, instead of 30/100 as in the gold reasoning trace, causing the backtracking search to fail. The arrows indicate what the reasoning trace looks like when assuming the 30 is representing 30%.

Figure 14: Coconut + GPT-2 Small’s vocabulary projections, from instance 460 of GSM8k-Aug’s test split. The gold reasoning trace is encoded, except for the intermediate result 21, which causes the backtracking search to fail.

Figure 15: Percent of any gold reasoning trace found in the vocabulary projections of latent tokens for correctly answered problems by reasoning trace length. Reasoning traces with more than 5 steps are not shown due to low instance counts (

<3

Method	Base Model	Incorrect Instances	Correct Answer in Top-10	Percent
Coconut	GPT-2 Small	793	369	46.5%
CODI	GPT-2 Small	675	337	49.9%
Coconut	Llama-3.2-1B-Instruct	759	425	56.0%
CODI	Llama-3.2-1B-Instruct	511	275	53.8%

Appendix

Table 14: Incorrect predictions on GSM8k-Aug where the correct answer appears in the top-10 predicted tokens.

Figure 16: CODI + Llama-3.2-1B-Instruct’s vocabulary projections, from instance 290 of GSM8k-Aug’s test split. The reasoning trace found and verified by the forward chaining method is displayed. This reasoning trace happens to match the gold reasoning trace.

Figure 17: Verification process for latent token 2, candidate step 1 from instance 290 of GSM8k-Aug’s test split.

Figure 18: Verification process for latent token 2, candidate step 2 from instance 290 of GSM8k-Aug’s test split.

Figure 19: Instance 1 of the train split of PrOntoQA shown as a directed acyclic graph. This instance is also shown in text in Figure 7 .

Implicit Reasoning in Language Models

Implicit reasoning in language models studies how models solve problems through internal computation rather than, or alongside, written chains of thought.

See all 95 papers

Observable Patterns Are Not Explanations: A Causal-Geometric Analysis of Latent Reasoning Models
Jun 10, 2026Darpan Aswal, Thomas Palmeira Ferraz, Yongxin Zhou +1Continuous Latent Reasoning Causal Interpretability
Université Grenoble Alpes, CNRS, Grenoble INP, LIG · Université Paris-Saclay · NAVER LABS Europe
Unlocking the Black Box of Latent Reasoning: An Interpretability-Guided Approach to Intervention
May 31, 2026Shuochen Chang, Tong Bai, Xiaofeng Zhang +5LLM Interpretability Causal Interventions in Language Models
1Shanghai Jiao Tong University · 2Fudan University
SPOT: Span-level Pause-of-Thought for Efficient and Interpretable Latent Reasoning in Large Language Models
Mar 6, 2026Yunlong Chu, Minglai Shao, Yuhang Liu +4Efficient Language Model Reasoning CoT Compression
School of New Media and Communication, Tianjin University, Tianjin, China · Independent Contributor, CA, USA · School of Computer Science and Engineering, Beihang University, Beijing, China

Jun 10, 2026J/K move · Enter open · S save

Observable Patterns Are Not Explanations: A Causal-Geometric Analysis of Latent Reasoning Models

Darpan Aswal, Thomas Palmeira Ferraz, Yongxin Zhou, Maxime Peyrard

Université Grenoble Alpes, CNRS, Grenoble INP, LIG · Université Paris-Saclay · NAVER LABS Europe

Continuous Latent Reasoning Causal Interpretability Large Reasoning Models Latent Reasoning Latent CoT Reasoning

Open paper AskPDF arXiv

Latent reasoning models (LRMs) replace explicit chain-of-thought with continuous thoughts. Recent work treats observable latent-state patterns, such as BFS-like frontiers and decodable arithmetic computation, as evidence for internal reasoning mechanisms. Evaluating two LRMs (Coconut and CODI) against controls lacking the proposed recurrence or curriculum, we find these patterns also appear in the controls and do not always causally affect behavior. Causal interventions reveal that latent-thought utilization is not binary but graded, scaling with a thought's causal effect on model behavior. Geometric analyses reveal this effect concentrates in low-rank directions whose step-to-step geometry grows more structured as their behavioral influence increases. Latent thoughts should therefore be treated as hidden computation, not hidden explanation: decodability, attention, or static structure alone cannot establish mechanism. LRM interpretability thus requires matched controls and causal tests.

Abstract

Implicit Reasoning in Language Models

Explore similar work

Observable Patterns Are Not Explanations: A Causal-Geometric Analysis of Latent Reasoning Models

Unlocking the Black Box of Latent Reasoning: An Interpretability-Guided Approach to Intervention

SPOT: Span-level Pause-of-Thought for Efficient and Interpretable Latent Reasoning in Large Language Models