Internalization

Large language model (LLM) agents operate in dynamic environments where knowledge continuously evolves. Existing memory systems typically treat external memory as a monotonically growing repository, inevitably leading to retrieval degradation and increasing computational costs over time. We argue that the core challenge is not retrieval alone, but managing the knowledge lifecycle: deciding what to externalize, update, or ultimately internalize. Inspired by Complementary Learning Systems (CLS) theory in neuroscience, we propose Dual-Layer Agentic Memory, a framework that shifts memory management to the write phase through cost-aware epistemic routing and periodic parametric consolidation. Incoming information is categorized as non-write, write-new, or write-update, and routed through a small-to-large model cascade that minimizes routing overhead while filtering redundant memories. A subsequent write-back phase selectively consolidates high-value external memories into model parameters via supervised fine-tuning. Experiments demonstrate the dual efficiency of our approach: a 1.7B/8B cascade prunes up to 68% of redundant external memory while escalating fewer than 50% of inputs, yet retains over 98% of the downstream QA Exact Match (EM) achieved by an exhaustive retention baseline. We further show that periodic consolidation successfully internalizes external knowledge, allowing the router to adaptively suppress redundant writes as the model's epistemic boundaries evolve. Overall, our framework presents a unified paradigm for agent memory: selective externalization followed by selective internalization. Code and dataset will be released upon acceptance.

Large language models are increasingly trained and deployed with long contexts that span documents, code repositories, and interaction histories. This scaling reflects the implicit assumption that training on longer contexts will only help the model by exposing it to richer evidence. We challenge this view by studying how the context window shapes a model's mode of learning, shifting it between parametric internalization and contextualization. We propose the Information Abundance Paradox, which hypothesizes that abundant relevant information in the training context can reduce the incentive to encode that information parametrically, thereby increasing reliance on context. In pretraining with long documents, increasing the context window improves language modeling, natural language understanding, and closed-book MCQA only up to an intermediate optimum, after which performance consistently declines. In supervised fine-tuning, more task-relevant train-time context improves performance with supporting context, but reduces robustness when context is absent or misleading at test time. Our analysis suggests that this behavior arises when longer context provides a lower complexity solution. Mechanistically, training with informative context shifts gradient pressure from feed-forward networks, often linked to parametric knowledge, toward attention modules, and causal interventions show that this shift increases reliance on context during inference. Overall, these findings support the Information Abundance Paradox and suggest that scaling toward near-infinite context is not simply a matter of supplying more data, even when high-quality long-context data is abundant.

Real-world multimodal instructions often bundle multiple requirements with unequal importance, yet most multimodal training data still reduce instruction following to answering one self-contained question. We study this gap through rubric comprehension, which casts the model not as a generator measured against rubrics but as an executor that follows them: given an image and a typed, prioritized rubric, the model must verify each rule before producing an overall judgment. To support this setting, we propose PRISM, a four-stage data synthesis framework that produces persona--task pairs, prefix-guided rule sets, quality-filtered rubrics, and structured verification traces. We further introduce PRISM-Eval, whose Loose and Strict metrics use deterministic matching against fixed labels and therefore require no inference-time judge model. With only 10K synthesized samples, PRISM lifts Qwen3-VL-4B from 9.5% to 30.1% Strict accuracy on PRISM-Eval while preserving average performance on general benchmarks, and the gains transfer to four additional open-source MLLMs across dense and MoE architectures, suggesting that structured rubric supervision is a scalable path toward multi-rule, priority-aware multimodal instruction following.

Recent work on LLM agents is shifting from external capability elicitation to capability internalization, enabling agents to retain useful skills without retrieval at inference time. On-policy self-distillation (OPSD) offers a promising direction, but many existing methods typically supervise students by scoring actions along student-generated trajectories. Such supervision has two limitations: teacher preferences are not validated by environment outcomes, and action-level scores underuse information from student rollouts, teacher rollouts, and their behavioral relationship. We therefore advocate outcome-verified teacher supervision and comparative learning over teacher-student trajectories. Based on this view, we propose Outcome-Verified Comparative Self-Distillation (OVCSD). OVCSD organizes failed student rollouts into a prefix tree, adaptively invokes a skill-conditioned teacher from student-reached states, and retains only outcome-verified successful continuations. It then applies localized comparative learning at the first state-aligned divergence and distills the post-divergence teacher suffix to transfer completion behavior. Experiments on ALFWorld and WebShop across three model scales show that OVCSD consistently outperforms skill-free RL and existing self-distillation baselines, achieving up to 29.7 and 5.4 absolute success-rate gains over the strongest baselines on ALFWorld and WebShop, respectively, while adding less than 3% privileged interaction during training.

As large language models move from isolated task solving toward long-term service in human environments, they require social intelligence: the ability to infer mental states, track social relations, reason over norms, and adapt behavior under context. This report presents Zhijing, an integrated framework for measuring, internalizing, and grounding social intelligence. For measurement, we introduce SoMBench, a psychology-grounded benchmark spanning 3 primary dimensions, 17 secondary dimensions, and 71 task paradigms. It controls question format, narrative perspective, and context length across 284 shared scenarios and 3,481 expert-verified instances. Evaluation of 20 representative LLMs reveals substantial headroom: the best model achieves only 72.08% overall accuracy, and none of the 17 secondary dimensions reaches the 90% near-ceiling band. For internalization, we develop Zing, a diagnosis-driven training recipe combining supervised fine-tuning, on-policy distillation, and rubric-based reinforcement learning. Across five social-cognition benchmarks, Zing consistently outperforms its base models, with Zing-27B-Stage2 achieving the best average score and Zing-32B-Stage2 remaining competitive with DeepSeek-V4-Pro. For deployment-time grounding, we build Actio, a harness-controlled inference architecture that routes four typed supports into reasoning: PRISM for procedural guidance, Starling for runtime mental-state representation, SAGE for reusable experience, and gated RAG for external social and normative knowledge. Across five base models and three benchmarks, the full harness improves 14 of 15 model-benchmark pairs and is best or tied for best in 8, demonstrating the effectiveness of typed runtime support. Together, these results show that socially intelligent LLMs require coordinated advances in evaluation, parametric internalization, and deployment-time grounding.

Omni models transcribe clean, single-speaker speech well, but their accuracy drops sharply when speakers overlap and the scene is noisy, exactly where knowing who said what matters most. A natural fix is a short scene description. We show why this is risky: answer-bearing text lets the model copy instead of listen, so the score rises although nothing has been heard; a silent test exposes this shortcut at once. We call this failure mode perception bypass and address it with Audio-Grounded Scaffold Context (AGSC). AGSC links three steps: first, we build clues from audio to guide listening without giving the answer; second, answer-overlap and silence tests probe them for leakage and audio dependence; finally, those clues scaffold training but vanish at test time, yielding no-clue capability. Across three heterogeneous Omni models, training on AGSC lowers no-clue capped mean permutation word error rate (mpWER) on overlapping, noisy speech from 25%-71% to 9%-15%. For streaming control, we formulate a joint GDPO task in which the model learns when to use a clue and how to produce a speaker-attributed transcript from separately normalized format, gate, and transcript rewards. After internalization, AGSC adds almost no inference overhead.

We study internalization processes, by which neural-network-based systems absorb an explicit computational procedure into their own weights, and how they facilitate learning. We investigate how transformers internalize the simulation of semiautomata by internalizing chain-of-thought (CoT) tokens, which classes of semiautomata are harder to internalize, and expose the flip side of internalization, that is, a progressive degradation of out-of-distribution performance. We then provide the first provable analysis of successful internalization: for the task of learning parities, we show that a simplified one-layer transformer provably first learns the target with explicit CoT supervision and then internalizes the autoregressive generation as CoT tokens are progressively removed, learning to directly compute the parity. This task is computationally hard to learn from data without CoT supervision. Finally, we discuss how learning through internalization relates to the \textit{Positive Distribution Shift} phenomenon recently introduced by~\citet{Med+26}.

Generative listwise ranking with Large Multimodal Models (LMMs) aims to capture global list context in a single forward pass, but its effectiveness degrades in long-context multimodal scenarios. We identify a recurring failure mode, parse collapse, where the autoregressive decoder produces fluent yet incomplete rankings by silently omitting candidates and terminating early. This failure stems from limited context utilization rather than simple formatting mistakes, making prompt engineering and constrained decoding insufficient. We propose PRISMR (Parameterized Representation Internalization for Semantic Multimodal Ranking), a framework that replaces transient in-context list processing with parametric structural conditioning. PRISMR uses a lightweight hypernetwork to encode multimodal candidates in parallel and generate item-specific LoRA weights, which are synthesized into an instance-specific adapter for a LMM. This paradigm enables more robust internalization of list structure while preserving the base model. We further introduce a large-scale multimodal review-ranking benchmark for evaluation. Experiments demonstrate that PRISMR substantially reduces parse collapse, improves listwise ranking performance, and transfers effectively across domains and instruction-tuned backbones.

Recent work has shown that on-policy distillation can internalize privileged context, such as system prompts or task hints, into a student model so that the context is no longer needed at inference time. Although this approach successfully improves the student's no-context performance, we identify an interesting and previously unstudied phenomenon: in many settings, reintroducing the original privileged context to the distilled student actually degrades its performance, even on instances it already solves correctly without context. We term this context-induced degradation and argue that robust internalization demands not only matching the teacher's context-conditioned behavior, but also remaining stable when the context is reintroduced, a property we call context removability. Motivated by this observation, we propose a lightweight consistency regularizer that first anchors the student's no-context output via stop-gradient, then penalizes the context-conditioned output for deviating from it via forward KL divergence. This simple addition requires only one extra forward pass per training step, yet it effectively mitigates context-induced degradation and, in many cases, even improves no-context performance. Across 12 configurations spanning diverse domains and model families, our method improves context-conditioned accuracy in the majority of settings, reduces context-induced harm in 11 out of 12 settings, and effectively eliminates response-length inflation. A mechanistic case study further confirms that context removability is achieved at the representation level, with hidden states remaining nearly identical regardless of whether the context is present.

Language models can state that "the Earth orbits the Sun" and, when role-playing Aristotle, assert the opposite. Recent work argues that persona adoption is fundamental to how language models behave, with models selecting the most appropriate persona for a given context. Does such role-playing merely change the model's outputs, or does it also affect what the model internally represents as truthful? We study this question using the role-play of characters whose beliefs differ from the modern consensus, and induce personas with a number of different methods: prompting, in-context learning (ICL), supervised fine-tuning (SFT), and Open Character Training (OCT), and Emergent Misalignment (EM). We measure belief internalization across these approaches with truth probes and with behavioral tests, finding a broad spectrum of belief internalization. Prompting, ICL, and SFT change what the model says with little representational change. EM creates a large, broad shift in the model's truth representation, and OCT a smaller shift that is clearest on the larger model. Understanding when training changes a model's worldview rather than merely its behavior may become increasingly important as AI systems are entrusted with greater autonomy and influence.

Lifelong learning is essential for Large Language Model (LLM) agents operating in dynamic, interactive environments. However, existing lifelong learning agents for long-horizon tasks typically depend on discrete skill or past experiences retrieval with static parameters during inference, which prevents them from continuously internalizing test-time feedback like human learners. To bridge this gap, we propose Skill-enhanced Test-Time Co-Evolution (\texttt{LifeSkill}), a two-stage reinforcement learning framework for Online Lifelong Learning Agents. Specifically, we design Verifier-Guided Skill Learning that addresses the lack of direct supervision for skill extraction by rewarding candidate skills according to the average verifier success of multiple skill-conditioned policy rollouts, encouraging the model to generate skills that are useful for solving tasks rather than merely plausible in text. Furthermore, we introduce Online Skill Internalization, which continuously improves the policy model during test-time interaction by transforming skill-conditioned trajectories into reward signals. This enables the agent to directly internalize reasoning capabilities into its parameters, avoiding the context bloat of experience retrieval. Experiments on LifelongAgentBench show that LifeSkill improves average performance by 7 absolute points by comparing with existing lifelong agent baselines.

Experience internalization converts contextual experience from past interactions into reusable parametric capability, offering a promising path toward continual learning in large language models (LLMs). While prior work has predominantly focused on single-iteration transfer, we discover that under multi-iteration experience learning, existing methods suffer from a progressive capability collapse rather than compounding improvement. We systematically examine this failure through three vital dimensions of experience internalization: (1) Experience Granularity: We find that principle-level experience is more durable than instance-level experience, as it effectively abstracts transferable strategies away from trajectory-specific details. (2) Experience Injection Pattern: Our analysis reveals that step-wise injection significantly outperforms global injection by aligning experience with intermediate decision states, a property that is critical for long-horizon tool use. (3) Internalization Regime: We demonstrate that off-policy context-distillation on high-quality teacher trajectories provides a substantially more stable training signal than on-policy context-distillation, which is inherently limited by local corrections on student-induced flawed states. Together, these insights yield a simple yet robust recipe for stable and sustainable experience internalization, providing concrete guidance for engineering self-evolving and continually learning LLMs.

Long-horizon LLM agents can benefit from reusable skills, yet existing skill-based methods often rely on external skill generators during training or persistent skill retrieval at inference, increasing engineering complexity, context length, and deployment latency. We propose Self-Internalizing Reinforcement learning with Intrinsic skills (SIRI), a three-phase framework that enables agents to discover, validate, and internalize skills without external skill generators or inference-time skill banks. SIRI first warms up the policy with GiGPO to acquire basic interaction ability and collect successful skill-free trajectories. It then performs self-skill mining, where the current policy summarizes compact skills from its own successful plain rollouts and validates them through paired skill-augmented and skill-free rollouts. Finally, SIRI distills only beneficial skill-guided action tokens into the plain policy using trajectory-level utility and action-level advantage. At inference, the agent runs with the original prompt only. On ALFWorld and WebShop with Qwen2.5-7B-Instruct, SIRI improves GiGPO from 0.908 to 0.930 on ALFWorld and from 0.728 to 0.813 on WebShop, outperforming prompt-based, RL-based, and memory-augmented baselines. Further analysis shows that our self-mining strategy can achieve performance comparable to distillation with closed-source large model. Our code is available at https://github.com/kirito618/SIRI.

Reinforcement learning from verifiable rewards improves the reasoning ability of large language models, but often suffers from entropy collapse, in which increasingly concentrated policies reduce rollout diversity and useful learning signals. Existing remedies either constrain the RL objective (e.g., entropy regularization) or adjust sampling temperature during rollout collection, but these interventions remain external to the model parameters. We propose Temperature-Scaled On-Policy Self-Distillation (TS-OPSD), a lightweight policy reheating method that internalizes the exploratory effect of temperature into model parameters. Starting from an entropy-collapsed RL checkpoint, TS-OPSD constructs a self-teacher by applying high-temperature scaling to the model's own logits, then distills the resulting smoother distribution back into the student. This policy reheating requires no external teacher, privileged data, or additional inference cost. Experiments on Qwen3-4B-Base and Qwen3-8B-Base show that policy reheating yields a stronger initialization for continued RL than both standard continued RL and rollout-level temperature reheating. Further analyses show that TS-OPSD mainly reduces output sharpness while preserving intermediate representations, top candidate sets, and reasoning capability. These results suggest that entropy restoration can serve as a simple post-collapse intervention for extending reasoning-oriented RL.

Despite the rapid progress of multimodal large language models in building Graphical User Interface (GUI) agents, their real-world task completion is fundamentally bottlenecked by a lack of world knowledge about GUI operations. Existing solutions typically rely on expensive multi-agent scaffolding or conventional post-training paradigms, such as Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL). However, post-training only allows agents to implicitly absorb world knowledge through action annotations or reward signals, leading to inefficient trajectory memorization rather than genuine comprehension. Therefore, an approach that enables explicit learning of this knowledge is imperative. To this end, we propose GUI-CIDER, a mid-training method that explicitly internalizes GUI world knowledge through Causal Internalization and Density-aware Exemplar Reselection. GUI-CIDER operates in three stages: (1) data synthesis, which distills static planning and dynamic causal knowledge from GUI trajectories into text; (2) exemplar reselection, which filters the corpus by rewarding causal structures and penalizing semantic redundancy; and (3) mid-training, where the refined data is used to embed the acquired knowledge. Extensive experiments on two GUI knowledge benchmarks and three task completion benchmarks demonstrate that GUI-CIDER consistently improves both the agent's understanding of GUI operations and its task success rates.The codes are available at https://github.com/Wuzheng02/GUI-CIDER.

Equipping large language models with explicit skills has emerged as a promising paradigm for enabling autonomous agents to solve complex tasks. Agent skills can be inherently divided into general skills for broad cognitive transfer and task-specific skills for dynamic execution. However, existing skill-based reinforcement learning (RL) methods typically force a rigid choice between full externalization, which incurs prohibitive context overhead, and full internalization, which risks overfitting and knowledge conflicts. To address this dilemma, we propose Skill0.5, a novel agentic RL framework that explicitly differentiates skill treatments by combining general skill internalization with task-specific skill utilization. Driven by a dynamic, difficulty-aware router, Skill0.5 streams tasks into distinct mastery tiers to apply tailored optimization strategies: it internalizes general skills via privileged distillation to build a cognitive foundation for hard tasks, while using diagnostic probing on easy tasks to penalize shortcuts and enforce specific skill utilization. Experiments on ALFWorld and WebShop demonstrate that Skill0.5 outperforms both memory-based and skill-based RL baselines, yielding performance improvements across both in-distribution and out-of-distribution scenarios.

Structured skill prompts improve exploration in long-horizon agentic reinforcement learning (RL). Skill-augmented RL methods retain external skills at inference, while skill-internalization RL methods withdraw them during training to enable autonomous performance. However, existing internalization approaches only use skill-helpfulness contrast for curriculum control, leaving the policy update unchanged and unable to distinguish skill-dependent from autonomous success. We propose SkillC, a framework based on Contrastive Skill Credit Assignment (CSCA) that converts this contrast into a direct learning signal for internalization. \textsc{SkillC} samples paired skill-injected and skill-free rollouts for tasks from active skill types within the same policy update, and injects their task-level contrast into optimization via a dual-stream advantage estimator that preserves global ranking while applying a one-sided correction toward skill-free success. A smoothed validation-level signal further drives an adaptive curriculum over attribution strength, rollout allocation, and monotonic active-set pruning. Experiments on ALFWorld and WebShop show that, without runtime skill access, SkillC surpasses the strongest prior skill-internalization RL baseline by 5.5% and 4.4%, respectively, while remaining competitive with skill-augmented RL methods.

We present PEAM, a Parametric Embodied Agent Memory framework in Minecraft that transforms agent memory from inference-time retrieval into parameter-resident skills internalized through experience. PEAM pairs a slow deliberative LLM for open-ended reasoning with a fast parametric module for reflexive execution of consolidated skills. The fast module is a multimodal Mixture-of-Experts LoRA architecture with per-category physically isolated adapters, enabling parameter-level continual learning without catastrophic forgetting. We treat failure as a first-class training signal: failure--correction trajectory pairs are internalized through a joint behavioral-cloning and contrastive objective, so the agent learns not only what succeeds but also how corrected actions differ from failed ones. To govern consolidation, PEAM introduces a parameterization-worthiness score for deciding which experience should be internalized, and a scale-free self-triggered consolidation mechanism for deciding when to internalize without task-specific hand-tuned thresholds, making the agent self-evolving as the trigger transfers across task distributions without re-tuning. Experiments in Minecraft show that PEAM improves long-horizon task performance, mitigates forgetting on previously consolidated skills, and improves parametric-versus-retrieval efficiency over retrieval-based embodied agents and parametric memory variants.

Large language models are increasingly used as planning components in agentic systems, but current tool-use pipelines often require full tool schemas to be included in every prompt, creating substantial token overhead and limiting the practicality of smaller models. This paper investigates whether tool-use knowledge can be internalized into small language models through parameter-efficient fine-tuning, enabling structured planning without explicit tool descriptions at inference time. Using AssetOpsBench as the primary benchmark, we fine-tune Gemma 4 E4B and Qwen3-4B with 8-bit QLoRA on approximately 1,700 tool-use examples spanning tool knowledge, question-to-plan mappings, and execution-style traces. We evaluate the resulting models under description-free inference, where the prompt omits the tool catalog entirely. The fine-tuned models outperform an informed unfine-tuned baseline that receives full tool descriptions, reducing input length by 82.6% while improving structural and LLM-judge planning scores. In the best Gemma run, the model achieves an AT-F1 of 0.65 and an overall judge score of 3.88, compared with 0.47 and 2.88 for the informed baseline. Qwen3-4B achieves a strong overall judge score of 3.78 while using 62% less memory and running 2.5$\times$ faster than Gemma, though it also exhibits greater catastrophic forgetting on general multiple-choice benchmarks. Additional ablations show that LoRA rank controls a quality--retention trade-off, with $r=32$ maximizing planning quality and smaller ranks preserving more general knowledge. These results suggest that, for fixed tool catalogs, QLoRA fine-tuning can shift tool knowledge from prompt context into model weights, substantially reducing inference overhead while maintaining or improving tool-planning quality.

In LLM Reinforcement Fine-Tuning (RFT), curriculum learning drives both efficiency and performance. Yet, current methods externalize curriculum judgment via handcrafted heuristics or auxiliary models, risking misalignment with the policy's training dynamics. In this paper, we introduce METIS (METacognitive Internalized Self-judgment), a novel framework that internalizes curriculum judgment as a native capability. Leveraging a critical observation that within-prompt reward variance effectively gauges prompt informativeness, METIS predicts this metric based on recent training outcomes as lightweight in-context learning examples. This intrinsic self-judgment then dynamically dictates the training allocation. Moreover, METIS closes the loop between judgment and optimization by jointly optimizing the standard RFT rewards and a self-judgment reward. This allows the policy to learn what to learn next, as a form of metacognition. Across extensive discrete and continuous RFT benchmarks from mathematical reasoning, code generation, to agentic function-calling, METIS consistently delivers superior performance while accelerating convergence by up to 67%. By bypassing handcrafted heuristics and auxiliary models, our work establishes a simple, closed-loop, and highly efficient curriculum internalization paradigm for LLM reinforcement fine-tuning.

LLM-powered multi-agent systems can now automate the full research pipeline from ideation to paper writing, but a fundamental question remains: automation for whom? Researchers operate under different resource configurations, hold different methodological preferences, and target different output formats. A system that produces uniform outputs regardless of these differences will systematically under-serve every individual user, making personalization a precondition for research automation to be genuinely usable. However, achieving it requires three capabilities that current systems lack: accumulating reusable procedural knowledge across projects, retaining user-specific experience across sessions, and internalizing implicit preferences that resist explicit formalization. We propose NanoResearch, a multi-agent framework that addresses these gaps through tri-level co-evolution. A skill bank distills recurring operations into compact procedural rules reusable across projects. A memory module maintains user- and project-specific experience that grounds planning decisions in each user's research history. A label-free policy learning converts free-form feedback into persistent parameter updates of the planner, reshaping subsequent coordination. These three layers co-evolve: reliable skills produce richer memory, richer memory informs better planning, and preference internalization continuously realigns the loop to each user. Extensive experiments demonstrate that NanoResearch delivers substantial gains over state-of-the-art AI research systems, and progressively refines itself to produce better research at lower cost over successive cycles.

Multi-agent debate has been shown to improve reasoning in large language models (LLMs). However, it is compute-intensive, requiring generation of long transcripts before answering questions. To address this inefficiency, we develop a framework that distills multi-agent debate into a single LLM through a two-stage fine-tuning pipeline combining debate structure learning with internalization via dynamic reward scheduling and length clipping. Across multiple models and benchmarks, our internalized models match or exceed explicit multi-agent debate performance using up to 93% fewer tokens. We then investigate the mechanistic basis of this capability through activation steering, finding that internalization creates agent-specific subspaces: interpretable directions in activation space corresponding to different agent perspectives. We further demonstrate a practical application: by instilling malicious agents into the LLM through internalized debate, then applying negative steering to suppress them, we show that distillation makes harmful behaviors easier to localize and control with smaller reductions in general performance compared to steering base models. Our findings offer a new perspective for understanding multi-agent capabilities in distilled models and provide practical guidelines for controlling internalized reasoning behaviors. Code available at https://github.com/johnsk95/latent_agents

Recent years have witnessed growing interest in applying Large Reasoning Models (LRMs) to Machine Translation (MT). Existing approaches predominantly adopt a "think-first-then-translate" paradigm. Although explicit reasoning trajectories significantly enhance translation quality, they incur prohibitive inference costs and latency. To address these limitations, we propose ReflectMT, a two-stage reflection internalization algorithm for machine translation that employs a "translate-first-think-later" paradigm. Our approach develops the model's "translate-reflect-refine" capability through reinforcement learning. In the first stage, we cultivate the model's capacity for high-quality reflection and refinement, thereby enhancing its semantic comprehension and task-specific knowledge. In the second stage, we train the model to internalize the knowledge acquired during reflection. As a result, during inference, ReflectMT operates in a direct translation mode, producing high-quality translations on the first attempt without any explicit reasoning steps. Experimental results on datasets such as WMT24 demonstrate that our model's first-pass translations during inference outperform multi-step reasoning LRMs such as DeepSeek-R1 in both automatic metrics and GPT-based evaluation, achieving a 2.16-point improvement in GPT-based translation quality evaluation while reducing token consumption by 94.33%.