> For the complete documentation index, see [llms.txt](https://docs.deryve.xyz/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.deryve.xyz/ai-agent/model-guide.md).

# Model guide

Every model is pre-trained for Deryve’s Robinhood Chain perpetual-trading context. This includes Lighter DEX mechanics, funding rates, USDC collateral, and supported markets.

Choose a model by the speed and depth your strategy requires. Access depends on your [model tier](/ai-agent/model-access-and-tiers.md).

### Speed and depth

* **Fast:** under one second. Best for high-frequency polling.
* **Balanced:** one to three seconds. Best for routine daily trading.
* **Deep:** three to eight seconds. Best for complex strategies.

{% hint style="info" %}
A deeper model can improve complex analysis. It does not guarantee trade performance. Every decision still passes through Deryve’s risk engine.
{% endhint %}

## Basic models

All users can access these nine models.

### Claude Haiku 4.5

**Anthropic · Fast**

**Best for:** High-frequency signal reads and rapid-cycle decisions.

* Lowest latency for real-time market scans.
* Produces reliable structured output.
* Works well for BTC and ETH momentum scalps.

**Avoid for:** Complex multi-variable strategies or low-liquidity stock-token markets.

### GPT-4o Mini

**OpenAI · Fast**

**Best for:** Fast pattern recognition on crypto perpetuals.

* Quickly reads candle patterns.
* Supports cost-efficient high-frequency polling.
* Handles short-term momentum analysis well.

**Avoid for:** Deep macro reasoning or multi-asset correlation analysis.

### GPT-4.1 Mini

**OpenAI · Fast**

**Best for:** Rapid directional calls on major crypto pairs.

* Builds on GPT-4.1 for sharper signal reads.
* Provides low-latency instruction following.
* Works well for BTC, ETH, and SOL momentum.

**Avoid for:** Nuanced risk assessment or stock-token perpetuals.

### Gemini 2.5 Flash

**Google · Fast**

**Best for:** Real-time price-action and momentum analysis.

* Produces dependable structured output.
* Reads trend-continuation signals well.
* Keeps latency low during live market scans.

**Avoid for:** Extended context analysis or deeply reasoned position sizing.

### Gemini 2.5 Flash Lite

**Google · Fast**

**Best for:** Maximum-throughput signal scanning.

* Offers very fast inference for signal cycles.
* Provides a solid baseline for simple trend reads.
* Fits automated high-frequency polling.

**Avoid for:** High-stakes entries or complex market conditions.

### DeepSeek V3.2

**DeepSeek · Balanced**

**Best for:** Cost-efficient market-structure analysis.

* Provides strong analytical reasoning at low cost.
* Aggregates multi-factor signals effectively.
* Fits mid-cap crypto perpetual signals.

**Avoid for:** Very fast decision cycles.

### Llama 4 Maverick

**Meta · Balanced**

**Best for:** Balanced signal reads with an open-source model.

* Provides strong open-source reasoning quality.
* Generalizes across crypto and stock perpetuals.
* Delivers competitive accuracy against proprietary models.

**Avoid for:** Highest-stakes positions requiring maximum reasoning depth.

### Mistral Large

**Mistral · Balanced**

**Best for:** Reliable multi-signal market analysis.

* Combines candle, volume, and trend signals.
* Supports structured analysis reliably.
* Produces disciplined JSON output.

**Avoid for:** Ultra-low-latency workflows or extended reasoning chains.

### Qwen 2.5 72B

**Qwen · Balanced**

**Best for:** Competitive reasoning at minimal cost.

* Offers 72B-parameter depth for analytical tasks.
* Follows instructions well for structured trade output.
* Fits crypto-perpetual momentum reads.

**Avoid for:** Frontier-level reasoning or complex stock-token analysis.

## Agent Premium models

These 17 models require at least **1,000,000 $DYVE** tokens. See [Model access and tiers](/ai-agent/model-access-and-tiers.md).

### Claude Sonnet 4.5

**Anthropic · Balanced**

**Best for:** Balanced quality and speed across frequent trade decisions.

* Understands market structure better than Haiku.
* Calibrates risk and reward effectively.
* Includes funding-rate awareness in its rationale.

**Avoid for:** The most complex multi-asset strategies.

### Claude Sonnet 4.6

**Anthropic · Balanced · Recommended**

**Best for:** Most Deryve strategies.

* Adds improved analytical depth.
* Calibrates stop-loss and take-profit levels well.
* Handles crypto and stock-token perpetuals effectively.

### Claude Opus 4.7

**Anthropic · Deep**

**Best for:** Complex multi-factor perpetual strategies.

* Provides maximum Anthropic reasoning depth.
* Analyzes multi-asset correlations effectively.
* Fits complex stock-token perpetual analysis.

**Avoid for:** Fast signal cycles.

### Claude Opus 4.8

**Anthropic · Deep**

**Best for:** High-stakes positions requiring maximum reasoning depth.

* Provides deep context understanding.
* Fits OPENAI-PERP, ANTHROPIC-PERP, and low-liquidity perpetuals.
* Combines macro and on-chain signals effectively.

**Avoid for:** High-frequency or latency-sensitive workflows.

### GPT-4o

**OpenAI · Balanced**

**Best for:** Proven multimodal market analysis.

* Performs well on crypto-perpetual analysis.
* Recognizes chart patterns from price data.
* Calibrates confidence conservatively.

**Avoid for:** The deepest multi-step reasoning tasks.

### GPT-4.1

**OpenAI · Balanced**

**Best for:** Market-structure and on-chain analysis.

* Improves on GPT-4’s analytical quality.
* Aggregates multiple signals effectively.
* Fits crypto and stock-token perpetuals.

**Avoid for:** The most complex macro-positioning decisions.

### GPT-5

**OpenAI · Deep**

**Best for:** Nuanced macro positioning.

* Provides deep analytical capability.
* Handles funding-rate-aware position sizing.
* Fits stock-token perpetuals such as TSLA, NVDA, and AAPL.

**Avoid for:** High-frequency polling.

### GPT-5.1

**OpenAI · Deep**

**Best for:** Cross-asset signal analysis with macro awareness.

* Supports multi-market correlation analysis.
* Weighs BTC dominance against alt-perpetual conditions.
* Helps structure advanced risk and reward.

**Avoid for:** Simple, time-sensitive momentum entries.

### GPT-5.2

**OpenAI · Deep**

**Best for:** Advanced on-chain and macro trade decisions.

* Offers frontier analytical depth.
* Fits correlated stock-token and crypto-perpetual setups.
* Integrates multiple timeframes effectively.

**Avoid for:** Fast scanning workflows.

### o3

**OpenAI · Deep**

**Best for:** Complex multi-variable perpetual strategies.

* Supports detailed trade-logic analysis.
* Fits complex multi-leg position sizing.
* Handles high-stakes SOL, XRP, and DOGE analysis.

**Avoid for:** Low-latency trade decisions.

### o3 Mini

**OpenAI · Balanced**

**Best for:** Step-by-step reasoning at balanced speed.

* Reduces latency compared with o3.
* Fits mid-complexity strategies.
* Supports structured risk-management analysis.

**Avoid for:** Maximum-depth multi-asset decisions.

### o4 Mini

**OpenAI · Balanced**

**Best for:** High-stakes entry decisions with balanced speed.

* Applies modern reasoning to stop-loss levels.
* Balances reasoning depth and latency.
* Fits protective trade planning from mark prices.

**Avoid for:** The most complex portfolio-wide strategies.

### o4 Mini High

**OpenAI · Deep**

**Best for:** The most complex position-sizing decisions.

* Applies the highest reasoning effort in the o4 Mini series.
* Fits multi-asset perpetual baskets.
* Supports high-leverage position sizing.

**Avoid for:** Frequent real-time polling.

### Gemini 2.5 Pro

**Google · Deep**

**Best for:** Long-context, multi-signal analysis.

* Processes many candles in one context window.
* Correlates funding rates with price action.
* Fits comprehensive stock-token perpetual analysis.

**Avoid for:** Low-latency signal scans.

### DeepSeek R1

**DeepSeek · Deep**

**Best for:** Complex multi-step entry and exit logic.

* Produces detailed rationale for decisions.
* Handles complex crypto-perpetual setups.
* Supports trades with multiple conditions.

**Avoid for:** Fast reaction trades.

### DeepSeek V4 Pro

**DeepSeek · Balanced**

**Best for:** Professional-grade trade reasoning at competitive cost.

* Serves as DeepSeek’s flagship analytical model.
* Applies strong multi-factor signal weighting.
* Balances depth and speed effectively.

**Avoid for:** Maximum-depth reasoning tasks.

### Grok 3

**xAI · Balanced**

**Best for:** Real-time market awareness with strong reasoning.

* Combines crypto sentiment with price action.
* Supports BTC, ETH, and SOL directional calls.
* Provides strong market intuition.

**Avoid for:** Extended, deeply structured multi-step analysis.

### Choose your default

Start with **Claude Sonnet 4.6** for general adaptive analysis. Choose a Fast model for rapid scans. Choose a Deep model for complex positions.

Compare models across similar markets and regimes. Review enough completed trades before changing your default.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.deryve.xyz/ai-agent/model-guide.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.