# Cura 1T: Specialized Model for Agentic Healthcare

> A human-gated self-evolution loop that searches the data mixture yields a 1T healthcare LLM matching or exceeding frontier models.

- **Source:** [arXiv](https://arxiv.org/abs/2607.15314)
- **Published:** 2026-07-21
- **Permalink:** https://picx.dev/p/AFbgub
- **Whiteboard:** https://picx.dev/p/AFbgub/image

## Summary

## Summary (Overview)

- Cura 1T is a 1-trillion-parameter healthcare-specialized LLM post-trained from the Kimi-K2.6 base using a **human-gated self-evolution loop** that treats the data mixture as the primary object of search.
- The model achieves **state-of-the-art or near-top performance** across five healthcare benchmark panels: MedAgentBench, HealthBench Professional, HealthBench Hard, MedXpertQA, and AgentClinic.
- The self-evolution loop automates the cycle of **planning, training, evaluating, and refining** the data mixture, using targeted synthetic data construction (e.g., tool-use trajectories, behavior calibration, knowledge injection, retention anchors) rather than a single generic medical-data update.
- Cura 1T improves over the base model by as much as **+0.159 on HealthBench Professional** and **+0.093 on MedAgentBench**, while preserving strong out-of-domain performance on AIME, GPQA-Diamond, and τ₂-Bench.
- The central contribution is a **repeatable training loop** that identifies missing capabilities from graded failures, synthesizes corrective data, and curates a mixture that avoids unintended forgetting.

## Introduction and Theoretical Foundation

The paper addresses the challenge of building a single LLM that can handle three core healthcare use cases:

1. **Patient care** – clinician- and patient-facing responses that follow physician guidelines.
2. **Clinical reasoning** – expert medical question answering across text and images.
3. **Healthcare agentic tasks** – multi-turn diagnostic workups, EHR tool use, and long-horizon workflows.

These tasks have **distinct failure modes** (rubric omissions, missing facts, brittle reasoning, premature diagnosis, malformed tool calls), and a narrow update for one task can degrade another. The authors argue that the primary bottleneck is not a training algorithm or hyperparameter configuration, but **data construction**: healthcare training signal is scarce, fragmented across guidelines/exams/patient interactions/images/EHR workflows, and often cannot be checked by a simple executor.

The theoretical foundation combines:
- **Low-rank adaptation (LoRA)** (Hu et al., 2021) for parameter-efficient training of a 1T model.
- **Self-distillation fine-tuning (SDFT)** (Shenfeld et al., 2026), which anchors updates to the model’s own on-policy trajectories to avoid damaging native reasoning patterns.
- **Self-evolution and auto-research** patterns (e.g., Self-Refine, Reflexion, autoresearch), but with the data mixture as the search object rather than prompts, code, or hyperparameters.

## Methodology

The training is built on a **thin adapter-training stack** over Kimi-K2.6 (256K context, text + vision input). The core innovation is the **human-gated self-evolution loop** (Figure 2 in the paper):

### Self-Evolution Loop Components

| Component | Role in the loop | Output |
|-----------|------------------|--------|
| **Plan** | Define target behavior, metrics, data recipe, hyperparameters | Training specifications |
| **Train** | Run SFT to verify mixture and hyperparameters, RL to improve policy, SDFT to consolidate | Candidate model |
| **Evaluate** | Run benchmark harnesses, collect graded trajectories and failure summaries | Failed trajectories and metrics |
| **Refine** | Categorize failures, synthesize targeted data, curate next mixture, validate rows | Validated data refinement |
| **Human review** | Approve plan, inspect validation, decide keep/revert/deploy | Decision |

### Training Stack
- **SFT** – low-cost prerequisite to verify mixture and hyperparameters.
- **RL** – reward-driven improvement stage.
- **SDFT** – final consolidation step. The SDFT objective is:

$$
\mathcal{L}(\theta) = D_{\text{KL}}(\pi_\theta(\cdot|x) \parallel \pi(\cdot|x,c)) = \mathbb{E}_{y \sim \pi_\theta(\cdot|x)}\left[\log \frac{\pi_\theta(y|x)}{\pi(y|x,c)}\right]
$$

where $x$ is the prompt, $c$ is privileged context (e.g., an intervened trajectory, reference behavior, verified knowledge), and $\pi_\theta(\cdot|x)$ is the student distribution. The student sees only the original prompt after training, which keeps updates anchored to the model’s own producible trajectories.

### Data Refinement Skills
The agent uses five specialized synthesis skills:
- **Retention Anchor**: Add examples of already-solved behaviors to prevent forgetting.
- **Reasoning Correction**: Edit failed reasoning traces and seed new questions.
- **Knowledge Injection**: Identify missing clinical knowledge, ground with retrieved documents, synthesize QA rows.
- **Behavior Calibration**: Compare desired vs. observed behavior, synthesize rubrics, generate conditioned responses.
- **Other**: Task-specific failures (e.g., agentic workflow errors).
- **Data Mixture Curation**: Merge, deduplicate, cap, and reweight the training set.

The loop runs separately for each capability and is kept or reverted based on evaluation. The final Cura 1T model is trained from the **consolidated mixture** accumulated across successful rounds.

## Empirical Validation / Results

### Healthcare Benchmarks

The paper evaluates Cura 1T on five benchmark panels. Below are the key tables.

**Table 1: Improvement from Kimi-K2.6 base**

| Benchmark | Base | Cura 1T | ∆ |
|-----------|------|---------|---|
| MedAgentBench | 0.847 | 0.940 | +0.093 |
| HealthBench Professional | 0.503 | 0.662 | +0.159 |
| HealthBench Hard | 0.222 | 0.368 | +0.146 |
| MedXpertQA | 0.569 | 0.655 | +0.086 |
| AgentClinic | 0.754 | 0.796 | +0.042 |

**Table 3: MedAgentBench task success**

| Model | Intervention | Overall | Decision |
|-------|-------------|---------|----------|
| Kimi-K2.6 | – | 0.883 | – |
| Round 1 | Tool-use | 0.943 | keep |
| Round 2 | Tool-use + retention | 0.967 | keep |
| Round 3 | Harness bug fix | 0.973 | keep |
| **Cura 1T** | Consolidated data mixture | **0.940** | release |
| Claude Opus 4.8 | – | 0.937 | – |
| GPT-5.5 | – | 0.894 | – |
| Gemini 3.1 Pro | – | 0.913 | – |

**Table 4: HealthBench rubric scores at T=1.0**

| Model | Intervention | Professional | Hard | Decision |
|-------|-------------|--------------|------|----------|
| Kimi-K2.6 | – | 0.503 | 0.222 | – |
| Round 1 | Behavior correction | 0.601 | 0.332 | revert |
| Round 2 | Clean behavior mix | 0.634 | 0.372 | keep |
| **Cura 1T** | Consolidated data mixture | **0.662** | **0.368** | release |

**Table 5: MedXpertQA split and overall pass@1 at T=1.0**

| Model | Intervention | Text | Multimodal | Overall | Decision |
|-------|-------------|------|------------|---------|----------|
| Kimi-K2.6 | – | 0.484 | 0.672 | 0.569 | – |
| Round 1 | Reasoning correction | 0.447 | 0.656 | 0.541 | revert |
| Round 1L | Reasoning correction, extended | 0.454 | 0.657 | 0.545 | revert |
| Round 2 | Knowledge injection + retention | 0.521 | 0.703 | 0.603 | keep |
| Round 3 | Mixture refinement | 0.560 | 0.728 | 0.636 | keep |
| Round 4 | Data mixture tuning | – | – | 0.440 | revert |
| **Cura 1T** | Consolidated data mixture | **0.600** | 0.722 | **0.655** | release |
| Claude Opus 4.8 | – | 0.562 | 0.710 | 0.628 | – |
| GPT-5.5 | – | 0.596 | **0.771** | 0.675 | – |

**Table 6: AgentClinic pass@1 by subset under tool-native protocol**

| Model | Intervention | MedQA | MedQA Ext | NEJM | NEJM Ext | Overall | Decision |
|-------|-------------|-------|-----------|------|----------|---------|----------|
| Kimi-K2.6 | – | 0.869 | 0.827 | 0.400 | 0.567 | 0.754 | – |
| Round 2 | Interactive trajectory + retention | 0.841 | 0.841 | 0.800 | 0.717 | 0.807 | keep |
| **Cura 1T** | Consolidated data mixture | **0.879** | **0.850** | **0.800** | 0.625 | **0.796** | release |
| Claude Opus 4.8 | – | 0.841 | 0.874 | 0.800 | 0.608 | 0.794 | – |
| GPT-5.5 | – | 0.832 | 0.808 | 0.467 | 0.358 | 0.684 | – |

### Out-of-Domain Evaluation

Cura 1T was evaluated on:
- **Reasoning**: AIME 2025, AIME 2026, GPQA-Diamond – performance on par with frontier models.
- **Agentic**: τ₂-Bench (airline, retail, telecom) – matches or surpasses reported scores, particularly on retail and telecom.

### Evolution Path Summary

The evolution map (Figure 3) shows that of the 11 attempted interventions (across all benchmarks), **7 were retained** and **4 were reverted** due to subset degradation, insufficient factual coverage, non-termination, or premature diagnosis. The final Cura 1T model uses the consolidated mixture from retained rounds.

## Theoretical and Practical Implications

- **Data mixture is the primary search object**: The central lesson is that healthcare specialization is a data-mixture problem before a hyperparameter-search problem. The self-evolution loop succeeded because it used evaluation failures as evidence for what data was missing.
- **Different failure modes require different data actions**: HealthBench needed behavior calibration (cleaner principle-rubric data), MedXpertQA needed knowledge injection with retention, MedAgentBench needed tool-use trajectories, and AgentClinic needed interactive trajectory training. A single generic medical-data update would not address these diverse failure modes.
- **Retention anchors are critical**: Adding examples of already-solved behaviors prevents targeted repairs from causing unintended forgetting. This was especially important for MedXpertQA, where reasoning correction alone degraded performance until retention anchors were added.
- **SFT as a screening step**: The use of SFT before RL/SDFT to verify mixture and hyperparameters at low cost is a practical design choice for large-model training.
- **Out-of-domain preservation**: The healthcare gains did not collapse general reasoning or agentic ability, suggesting that the self-evolution loop, with retention anchors, successfully avoids catastrophic forgetting.

## Conclusion

Cura 1T demonstrates a **repeatable, data-centric self-evolution loop** for building a healthcare-specialized LLM. The key contributions are:

1. A **human-gated loop** that plans, trains, evaluates, and refines the data mixture using specialized synthesis skills (reasoning correction, knowledge injection, behavior calibration, retention anchors).
2. **State-of-the-art results** on five healthcare benchmarks, with the largest gains from tool-use trajectories (MedAgentBench), cleaner behavior data (HealthBench), knowledge+retention (MedXpertQA), and interactive trajectories (AgentClinic).
3. **Preserved out-of-domain performance** on math, science, and general agentic benchmarks.
4. **Empirical evidence** that a single generic medical-data update is insufficient; the data mixture must be adapted to the specific failure modes of each task.

**Limitations**: The results are bounded by current compute, data, and training regime. Future work should broaden clinical behavior coverage, target long-horizon agentic tasks, and explore full-parameter updates.

**Safety and clinical use**: Cura 1T is a research model, not a medical service. Strong benchmark scores do not establish safety for unsupervised clinical use.

---

_Markdown view of https://picx.dev/p/AFbgub, served by PicX — AI-generated visual whiteboard summaries of research papers._