Default and Optimized Models
Every Awarded AI environment has an “Optimized” model. It’s the default in AI Chat, and our team picks and updates it as stronger models become available, so it’s a solid starting point for most tasks. The newest Optimized model is GPT-5.6 Terra, which replaced Gemini 3.1 Pro in October 2026. In our team’s testing it gave more detailed responses and handled tools like file search more reliably, and it now powers chat file search too. You can switch to a different model in the model selector whenever a task calls for it.
Security & Compliance by Deployment Environment
FedRAMP Customers: The "Optimized," or default, model is Google Gemini 3.1 Pro. ALL models are approved to process CUI/ITAR data. Google and Anthropic models run through Vertex AI for added compliance.
Commercial Cloud Customers: The "Optimized," or default, model is Google Gemini 3.1 Pro. Plan to NOT process CUI or ITAR documents or information in Awarded AI. Ask your CSM or AM for more info.
On-Prem / Self-Hosted Customers: The "Optimized," or default, model is GPT-4.1. ALL models available in a self-hosted environment are approved to process CUI/ITAR data.
Model Name | Model Type | Security Compliance | Primary Use Case | Key Characteristics & Recommendations |
Google Gemini Pro (2.5 or 3.1) | Reasoning | FedRAMP + On-Prem: | Your first choice for most proposal development tasks: RFP analysis, outlining, and drafting technical sections. | The go-to all-rounder. Excellent reasoning, large context window, and delivers detailed, concise output without "fluff." The best starting point for nearly all proposal work. |
Anthropic Claude 4 Sonnet (Large Context) | Reasoning | FedRAMP + On-Prem: | General writing, marketing content, or tasks where generating high volume of output is needed. | The creative writer. Great at writing with a 1M token context window, so you're less limited on input and output volume. Can be more verbose; output may feel more formulaic. A powerful second option for creative or marketing-focused tasks. |
OpenAI GPT-4.1 | Non-Reasoning | FedRAMP + On-Prem: | The fallback model. Large-scale data processing and generation where advanced reasoning is not a primary requirement. | The reliable workhorse. Massive context window, making it great for processing large amounts of information or large documents. Use it if other models are down or not delivering the results you need. |
Grok-3 & Grok-4 | Non-Reasoning | Commercial Cloud Only | Brainstorming, human-like interaction, and creative ideation sessions. | The brainstormer. Great for creative sessions, exploring ideas, and more conversational outputs. Only available to commercial cloud customers. |
Anthropic Claude 4.6 Opus | Reasoning | FedRAMP: | Complex multi-document analysis and high-stakes writing where reasoning depth directly changes output quality. | The deep analyst. Use ONLY when reasoning depth actually changes the output quality. Best for: complex multi-document synthesis with nuance; high-stakes writing where tone and strategy matter; ambiguous or conflicting requirements; novel or unusual solicitations. Available to commercial cloud and FedRAMP customers. |
⚠ Avoid These Models
Model | Reason to Avoid |
GPT-4O | Limited context window. Other available models will perform significantly better for most use cases. |
O3-mini | Limited context window. Other available models will perform significantly better for most use cases. |
Model Selector Cheat Sheet
If your task is... | Use this Model First | As a Backup |
Analyzing an RFP & Creating an Outline | Gemini Pro (2.5 or 3.1) | GPT-4.1 |
Drafting a Technical Proposal Section | Gemini Pro (2.5 or 3.1) | Claude 4 Sonnet (Large Context) |
Writing Marketing or Web Content | Claude 4 Sonnet (Large Context) | Grok-3 / Grok-4 |
Complex Multi-Document or High-Stakes Analysis | Claude 4.6 Opus | GPT-4.1 |
Experiencing issues with other models | GPT-4.1 | Claude 4 Sonnet (Large Context) |
Key Takeaways
Know What Model You're Using: Your "Optimized," or default, model varies by deployment environment (FedRAMP = Gemini 3.1 Pro | Commercial = Gemini 3.1 Pro | On-Prem = GPT-4.1).
Start with the Best: Use Gemini Pro (2.5 or 3.1) as your primary model for most proposal development tasks.
Use the Right Tool for the Job: Match the model's strengths to your specific task using the cheat sheet above.
Experiment! Your preferences matter. Try different models to see which works best for your style and workflow.
Decision Workflow
Know Your Environment
Determine if your organization is deployed on FedRAMP, Commercial Cloud, or On-Prem. This dictates your "Optimized," or default, model and whether CUI/ITAR data is safe to process.
Assess Your Task
For most proposal work (RFP analysis, outlining, drafting): start with Gemini Pro (2.5 or 3.1).
For creative writing or high-volume output: use Claude 4 Sonnet (Large Context).
For complex, multi-document analysis or high-stakes writing where nuance matters: reach for Claude 4.6 Opus (commercial cloud only).
For brainstorming or creative ideation: try Grok-3 or Grok-4 (commercial cloud only).
If other models aren't performing: fall back to GPT-4.1.
Use the Cheat Sheet
Reference the Model Selector Cheat Sheet above for quick task-to-model mapping.
Verify Outputs
Regardless of model, always verify that outputs are accurate for your specific purpose.
Index of Key Terms
Reasoning vs. Non-Reasoning Model
Reasoning Model: Designed to intuitively understand the underlying intent and context of a query. It can interpret complex requests, make logical inferences, and provide more intelligent, nuanced, and context-aware responses. Examples: Gemini Pro, Claude 4 Sonnet, Claude 4.6 Opus.
Non-Reasoning Model: Operates primarily by recognizing and replicating patterns from its training data. Highly effective for straightforward data processing and generation but does not possess the same level of inferential or interpretive capability. Examples: GPT-4.1, Grok.
Context Window
The total amount of information (both the user's input and the model's previous responses) that an AI model can access and "remember" at any given moment. Measured in tokens, a larger context window allows the model to handle longer documents and maintain consistency over extended interactions.
Token
The basic unit of data that AI models process. A token can be a whole word, part of a word, a number, or a piece of punctuation. For example, "government contracting" might be broken down into three tokens: "govern", "ment", and "contracting". Context window size is measured in the number of tokens it can hold.
CUI / ITAR Data
CUI (Controlled Unclassified Information): A category of information owned or created by the U.S. government that, while not classified, requires safeguarding and dissemination controls in accordance with federal laws, regulations, and government-wide policies.
ITAR (International Traffic in Arms Regulations): A set of U.S. government regulations that control the export and import of defense-related articles and services. Data subject to ITAR is highly sensitive information related to U.S. military and defense technologies, and it requires strict handling to protect national security.
DISCLAIMER: Regardless of the model you use, always verify the outputs are accurate for your purpose.