Xaira Therapeutics · Internal Strategy
Open Frontier Intelligence
for Drug Discovery
As general-purpose AI commoditizes, durable advantage shifts to
the deepest domain-specific scientific intelligence.
Prepared by Bo Wang · Chief AI Scientist
July 22, 2026
Competitive landscape updated July 22, 2026
Title slide. Set context: this is an internal strategy brief, not a product announcement. The landscape analysis was last updated today.
The strategic context
The AI platform shift
Yesterday's question
Can AI do this?
Foundation models proved they could reason, code, and synthesize knowledge. The answer is yes.
Today's question
Which layer holds the durable advantage?
General capability is rapidly commoditizing across both closed and open-weight models.
🧬
For drug discovery, the durable edge will not come from having the largest general model.
It will come from the deepest scientific intelligence — grounded in proprietary biological data,
experimental infrastructure, and expert feedback loops.
The base model is the commodity. The scientific intelligence built on top of it is the moat.
Key framing: general-purpose AI capability is becoming commoditized. The Anthropic/OpenAI race is already producing open-weight models at near-frontier levels. The question for Xaira is: where do we build? Answer: above the base-model layer.
Competitive landscape · Updated July 22, 2026
The open frontier is accelerating
| Model |
Developer |
Weights |
Context |
Key capability |
Relevance to Xaira |
| GLM-5.2 |
Z.ai (Zhipu AI) |
✓ MIT · Downloadable |
1M tokens |
Long-horizon reasoning, agentic tool use, coding |
Long-context analysis; scientific reasoning chains |
| Kimi K2.6 |
Moonshot AI |
✓ Open weights |
256K tokens |
Multimodal agentic; swarm orchestration (up to 300 agents) |
Complex multi-step task decomposition; vision input |
| Kimi K3 (announced) |
Moonshot AI |
⏳ Expected July 27, 2026 |
TBA |
3T-class; new attention architecture; frontier reasoning |
Cannot verify — weights not yet public |
| Inkling |
Thinking Machines Lab |
✓ Open weights (Acceptable Use) |
1M tokens |
Native text + image + audio; calibrated confidence; MoE 975B/41B active |
Multimodal biological data; uncertainty quantification |
| DeepSeek-V4-Pro |
DeepSeek |
✓ Open weights |
1M tokens |
Top open-source on knowledge and reasoning; 1.6T/49B active MoE |
Scientific literature synthesis; complex reasoning |
"The strategic question is no longer whether open-weight models will become capable enough.
It is how Xaira can turn rapidly improving general capabilities into proprietary scientific intelligence."
Important framing: GLM-5.2, Inkling, Kimi K2.6, and DeepSeek-V4-Pro are NOT drug-discovery models — they are evidence that the open frontier is moving fast. Kimi K3 weights were not public as of July 22, 2026 (expected July 27). Inkling has an Acceptable Use Policy, not pure MIT — license review needed. DeepSeek-V4-Pro license should also be reviewed for commercial use.
The architecture
Build above the base-model layer
Xaira's durable asset is the scientific intelligence built above the base-model layer
— not dependence on any single external model.
The base model is interchangeable. As open-weight models improve —
and they will — Xaira can swap in a better foundation without rebuilding the stack above.
Interchangeable · Base Layer
Open Frontier Model
GLM-5.2 · Kimi K3 · Inkling · DeepSeek-V4-Pro · future
Xaira — Post-training
Scientific Reasoning Layer
Domain-specific fine-tuning + RL from expert feedback
Xaira — Knowledge
Proprietary Biological Knowledge
Internal publications · program context · literature RAG
Xaira — Infrastructure
X-Cell · Models · Agents · Tools
Causal virtual cell · X-Atlas · experimental systems
Output
Therapeutic Decisions & Wet-Lab Outcomes
Closed loop: experiments validate and update the stack
X-Cell is Xaira's causal virtual cell model trained on X-Atlas (CRISPR perturbation data). The architecture is deliberately modular: the base model is the least proprietary layer. Everything above — post-training, biological knowledge, X-Cell grounding, agent tools, wet-lab validation — is Xaira's moat. Emphasize that this architecture allows model-swapping as better open-weight models emerge.
Internal capability assessment
Model evaluation framework
Propose an internal evaluation program comparing open-weight candidates on drug-discovery tasks.
Key principle: use proprietary Xaira scientific tasks — not public benchmarks alone.
🧬
Bio & chem knowledge
Domain knowledge depth
🔗
Causal reasoning
Perturbation → phenotype logic
📄
Long-context analysis
Multi-document synthesis
🖼
Multimodal understanding
Omics · imaging · structure
🧪
Experiment planning
Hypothesis → assay design
📊
Perturbation / omics
X-Atlas CRISPR data reasoning
🎯
Uncertainty calibration
Confidence vs. correctness
🔧
Tool use & agentic exec
Multi-step autonomous tasks
⚡
Latency · cost · privacy
License · fine-tune feasibility
Full evaluation dimensions from brief: biological/chemical knowledge, causal reasoning, long-context analysis, multimodal scientific understanding, experiment planning, perturbation/omics data interpretation, uncertainty calibration, tool use, agentic execution, latency/cost, deployment flexibility, fine-tuning feasibility, data privacy/security, license restrictions, robustness to changing evidence. Design evaluation tasks around actual Xaira scientific problems — not just public leaderboards.
Strategic options
Three strategic roles for open frontier models
1
Role 1 · Post-training
Base-model candidates
Start from an open-weight checkpoint and fine-tune on Xaira's scientific data.
Full control over the model's behavior, no API dependency.
Fine-tune feasible
Data privacy
2
Role 2 · Synthetic data
Teacher models
Use a capable open-weight model to generate and critique scientific reasoning traces,
without deploying it in production. Distill its knowledge into a smaller, Xaira-tuned model.
No inference cost
3
Role 3 · Orchestration
Orchestrated model portfolio
Dynamically route tasks to the best model based on capability, cost, modality, privacy, and latency.
No single permanent foundation model.
Resilient
Recommendation: evaluate whether Role 3 (portfolio orchestration) offers more resilience than committing to one permanent foundation model.
The open frontier is moving fast enough that today's best checkpoint may not be tomorrow's.
These three roles are not mutually exclusive. Role 1 and Role 2 could be combined: use a large open-weight teacher to generate training data, then fine-tune a smaller model as the production system. Role 3 is the most future-proof if the model landscape continues moving quickly. Consider starting with Role 2 (lowest deployment risk) to build the training data pipeline while evaluation for Roles 1 and 3 is underway.
Xaira's durable advantage
What transforms general intelligence into scientific intelligence
Domain-specific post-training
Fine-tuning on Xaira's scientific tasks and expert feedback
Proprietary multimodal biological data
Internal omics, imaging, structural, and experimental datasets
X-Cell grounding
Causal virtual cell model trained on X-Atlas CRISPR perturbation data
Retrieval over internal knowledge
Scientific publications, program history, clinical context
Specialized scientific tools
Structure prediction, docking, analysis pipelines
Agent orchestration
Multi-agent planning, long-horizon execution, tool calling
Expert feedback loops
Biologists, chemists, clinicians in the training loop
Wet-lab validation
Experimental outcomes close the learning loop
Continual learning from experimental outcomes
Every assay, every experiment updates the system — competitors without wet-lab infrastructure cannot replicate this
The key insight: none of these moat components are replicable by an AI lab releasing a general-purpose model. They require years of biological data collection, expert feedback, wet-lab infrastructure, and closed-loop experimental learning. The combination of X-Cell (causal virtual cell trained on X-Atlas) + expert feedback + wet-lab validation creates a compounding advantage that increases over time.
Call to action
Proposed next steps
Three concrete actions to start building Xaira's open frontier intelligence capability within 90 days.
1
Launch a structured internal model evaluation
Design a benchmark suite using real Xaira scientific tasks across the 9 evaluation dimensions.
Test GLM-5.2, Inkling, DeepSeek-V4-Pro, and Kimi K2.6 against each other on tasks
representative of Xaira's actual drug discovery workflow. When Kimi K3 releases (expected July 27, 2026), include it.
2
Designate a model strategy owner
Assign a senior technical lead to own the open-weight model strategy: evaluate candidates,
monitor license terms, track the model landscape, and recommend the architecture (base candidate, teacher, or portfolio).
This is a sustained function, not a one-time analysis.
3
Establish a 90-day pilot
Select one high-value drug discovery task. Deploy an open-weight model (Role 1 or 3).
Integrate with X-Cell, relevant retrieval systems, and at least one wet-lab feedback loop.
Treat the pilot as infrastructure investment, not a demo — the goal is the closed-loop learning system.
Decision needed: are we pursuing Role 1 (post-training), Role 2 (teacher), or Role 3 (portfolio)? Recommend the 90-day pilot clarifies this. The model strategy owner role is critical — without it, this becomes a project rather than a durable capability. Consider pairing with the AI team's existing infrastructure work around X-Cell.