DatBot blends

Available Models:

• Fast Blend: This is a mix of amazing models that return >100 tokens/second at a minimum from our partners, including Google’s Gemini 2.0 Flash and Meta’s Llama 4.
• HQ Blend: This is our slower, but incredibly high quality blend, including the best model out currently, Gemini 2.5 Pro.
• Thinking Blend: We include thinking models here, like o3-mini and DeepSeek’s R1, to ensure that different arachitectures collaborate on your answers.
• MoA: Our specialized reasoning mixture of agents. Not directly comparable to o3, it has its own place. You can think of it as several models discussing the right solution with one another, and then agreeing on the best answer.
• DB-1: Our Alpha attempt at long output reasoning - comparable to o1 in some ways, but has its own distinct uses given longer output potential.
• DB-1 Mini: Our Alpha attempt at long output reasoning - comparable to o3-mini in some ways, retains the incredibly long output.

Importance

DatBot Blends and MoA are, to my knowledge, the first time that “frontier” models have been put into a blended generation capacity in order to improve their performance. You can read more about this in the DatBot post about blends, here. No single origin here! DB-1 and DB-1 Mini are some of the first real successes at replicating o1’s long output, done independently in a way that’s cheap for end users.

OpenAI Models

Available Models:

Reasoning Models:

• o3: Expensive, but great. OpenAI’s newest and best reasoning model, with three sub-models - low, medium and high, which determine the reasoning effort that the model uses to solve your answer - in practice, low will be super quick, but I usually use medium or high.
• o4-mini: OpenAI’s pint-sized reasoning model. Same ‘reasoning effort’ options as o3.
• o3-mini: Kept for compatibility. Replaced by o4-mini.
• o1: Reasoning model built on GPT-4o.
• o1-mini: Smaller version of o1, largely replaced by o3-mini at this time.

Non-Reasoning Models:

• GPT-4.5: The most expensive model from ChatGPT
• GPT-4.1: Newer and better than GPT-4o, which ChatGPT uses.
• GPT-4o: The same model that ChatGPT uses.
• GPT-4o mini: Smaller, faster version of GPT-4o with great performance for its size.
• GPT-3.5 Turbo: The original ChatGPT model, used from writing and summarization to translation and coding assistance. The model that started the craze, and catapulted ChatGPT to fame! Similar in quality to our Instant Blend (probably a bit worse).
• GPT-4 Turbo: Much improved on the original GPT-3.5 Turbo, with better performance and all-around capabilities. This model has been updated consistently, and we use the latest version of it.

Importance of OpenAI:

OpenAI’s models are at the bleeding edge of AI technology, pushing the boundaries of what AI can accomplish. They proved what’s possible. Closely partnered with Microsoft.

Cohere Models

Available Models:

• Command-A: Cohere’s current top model, very good at responding in certain formats
• Command-R: Now outdated model, comparable to GPT 3.5-Turbo or a bit better.
• Command-R+: Stronger than the original GPT-4, Cohere’s last-gen “frontier” model.

Anthropic Models

Available Models:

• Claude 4 Opus: The new gold standard. ‘Nuff said. Also comes with a ‘thinking’ mode like OpenAI o1 and DeepSeek R1’s reasoning, the best coding model around in general.
• Claude 4 Sonnet: Just a smidge behind Claude 4 Opus, amazing model.
• Claude 3.7 Sonnet: Now replaced by Claude 4.
• Claude 3.5 Sonnet: Now replaced by Claude 4. Much stronger than original GPT-4, great price/performance ratio.
• Claude 3.5 Haiku: Lightning fast, better than GPT-4 in some tests, amazingly cheap.
• Claude 3 Opus: Fully replaced by Claude 4 Opus. Had a good run.

Importance:

The ‘Sonnet’ models tend to be the best price/performance - Sonnet 4 is an amazing coding model, for example (I use it all the time). Opus is often the best writer, at any price.

Meta Models

(In case you didn’t know… Meta owns Facebook, Instagram, Threads, WhatsApp, Oculus, etc. It’s just like Google is technically part of Alphabet alongside Waymo etc.)

Available Models:

• Llama 4 Scout: New price/performance champ, matching Gemini 2.0 Flash from Google
• Llama 4 Maverick: New best pound for pound contender - fast, reasonably priced, comparable to DeepSeek V3
• Llama 3.1 8b: Great price/performance ratio, matching GPT-3.5 Turbo quality but faster and cheaper. Largely the Gemini Flash series is a better price/performance champion now.
• Llama 3.3 70b: Pound-for-pound excellence, potentially as good as Llama 405b, according to Zuckerberg!
• Llama 3.1 405b: Meta’s flagship, comparable to GPT-4o or Sonnet 3.5.

Importance:

Meta is the standardbearer of quasi-open source AI (there are some limits to their license, but not meaningful unless you’re a many-billion dollar company).

Google Models

Available Models:

• Gemini 2.5 Pro: Best model in the world for many things here in 2025.
• Gemini 2.5 Flash: Google’s best high-speed model.
• Gemini 2.0 Flash: Google’s best price for the speed
• Gemini 2.0 Flash Lite: Super speedy
• Gemini 1.5 Pro: Google’s flagship model, competitive with top-tier offerings.
• Gemini 1.5 Flash: Optimized for speed while maintaining quality.
• Gemini 1.5 Flash 8b: Smaller, faster version of Flash.
• Gemma 3 27b: Open weights, similar to Llama 3.3 70b, very good model.
• Gemma 2 9b: Open weights model, great for specific tasks.
• Gemma 2 27b: Larger open weights model with broader capabilities.

Importance:

Google’s … well, Google. They make both closed source Gemini models that compete at the frontier, and open weight models (Gemma) anyone can run.

xAI Models

Available Models:

• Grok 3: xAI’s largest, competitive with ChatGPT 4.1 or Claude 4 Sonnet
• Grok 3 Mini: xAI’s answer to o3-mini from OpenAI
• Grok 2: X.AI’s older model, eclipsed by Grok 3.

Importance:

xAI and Elon Musk have a rivalry with OpenAI that’s heating up. With xAI buying X (formerly Twitter), they have a unique source of training data like Facebook or Google have, and they are investing rapidly in GPUs. Expect good models here.

Qwen Models

Available Models:

• Qwen 3 225b a22b: Alibaba’s largest model with 225 billion parameters.
• Qwen 3 32b: 32 billion parameter model in the Qwen 3 series.
• Qwen 3 30b a3b: 30 billion parameter alignment-tuned model.
• Qwen 3 14b: Medium-sized Qwen 3 model.
• Qwen 3 8b: Smallest model in the Qwen 3 series.
• Qwen 2.5 72b: Alibaba’s powerful multilingual model. Super solid model.
• Qwen 2.5 32b: Great for the price / performance.
• Qwen QwQ 32b: Alibaba reasoning model, cheap and comparable to R1.

Importance:

Alibaba’s Qwen models provide excellent performance across various sizes, with the Qwen 3 series representing their latest advancements in large language models.

Mistral AI Models

Available Models:

• Mistral Large: Mistral’s largest, competing at the frontier level with GPT-4o and Sonnet 3.5. Now outdated.
• Mistral Medium 3: Mostly replacing Mistral Large at a cheaper price.
• Mistral Small 24b v3.1 - super fast, very competitive, a real price/performance champ like Gemini 2.0 Flash.
• Ministral 8b: Fast and powerful 8B parameter model. Better than Llama 3.1 8b at most tasks.

Importance

Europe’s great hope for AI competitiveness, this French company is pretty neat.

Deepseek Models

• Deepseek Chat (V3): Great and inexpensive model from Deepseek, an amazing Chinese AI lab. Comparable to Sonnet 3.7, Llama 4 Maverick or GPT-4o. We use more expensive providers that do not train on output, instead of DeepSeek itself (which does train on your output).
• Deepseek R1: Reasoning model comparable to o1-preview. We use the more expensive providers that do not train on outputs, for privacy reasons.

Other Available Models:

• DeepSeek R1 Distill 70b - this is a Llama 3.3 70b distillation from R1, comparable to O1 but much cheaper, this isn’t a common model but it’s incredibly valuable.
• Hermes 405b: Steerable finetune of Llama 405B.
• Yi Large: Strong performer from 01-AI.
• Mythomax: Specialized creative writing and storytelling model.

Looking for non-LLM Integrations?

• We have a web scraper built in, and can scrape any website you want, either through our RAG knowledge map implementation - or through our chat interface.