DatBot Blends
- Fast Blend: A mix of amazing models that return >200 tokens/second at a minimum from our partners, targeting ChatGPT quality level.
- HQ Blend: Our slower, but incredibly high quality blend, including a few of the best models out at any given time.
- Thinking Blend: High quality thinking models from different frontier labs to ensure that different architectures collaborate on your answers.
- Cascade (Perspective Cascade): Multiple models exploring different perspectives, synthesizing a more valuable final output than a single model could.
- DB-1 (Deep Reasoning): Long output reasoning - able to tackle longer projects with more complex needs.
- DB-1 Flash (Deep Reasoning Flash): Blistering fast version of Deep Reasoning, matching long output reasoning with alacrity.
DatBot Blends and Perspective Cascade 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!
Deep Reasoning and Deep Reasoning Flash were some of the first real successes at generating stronger long output - now reasoning models are getting better at this natively, and this takes top models to still another level.
OpenAI Models
Thinking Models:
- GPT-5.1: OpenAI’s latest and most advanced thinking model, with adaptive reasoning that dynamically adjusts thinking time based on task complexity
- GPT-5.1 Codex: Full-size coding-focused thinking model, natively trained for agentic coding workflows
- GPT-5.1 Codex Mini: Smaller, faster coding-focused version of GPT-5.1
- GPT-5.1 Low: Low reasoning effort variant for faster responses
- GPT-5.1 High: High reasoning effort variant for complex problems
- GPT-OSS-120b: OpenAI’s open-source thinking model with 120 billion parameters
- GPT-5 Mini: Smaller, faster version of GPT-5
- GPT-5 Nano: Smallest GPT-5 variant, great for simple tasks
Non-Thinking Models:
- GPT-5.1 Chat (ChatGPT Version): Non-thinking variant of GPT-5.1, exactly the same as ChatGPT
- GPT-4o: The model that ChatGPT used to use (now outdated, use GPT-5.1 Chat instead)
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.
Anthropic Models
Thinking Models:
- Claude Opus 4.5 (Thinking): Claude’s most advanced model with thinking/reasoning capabilities
- Claude Sonnet 4.5 (Thinking): Sonnet 4.5 with thinking capabilities for complex reasoning
- Claude Haiku 4.5 (Thinking): Lightning fast thinking model, delivers similar coding performance to Claude Sonnet 4 at one-third the cost and more than twice the speed
Non-Thinking Models:
- Claude Opus 4.5: Non-thinking variant, excellent for creative writing
- Claude Sonnet 4.5: Non-thinking variant, excellent for coding and general tasks
The ‘Sonnet’ models tend to be the best price/performance - Sonnet 4.5 is an amazing coding model, for example (I use it all the time). Opus is often the best writer, at any price. Haiku 4.5 now offers remarkable capability at blazing speed - just six months ago, this level of performance would have been state-of-the-art.
Google Models
Thinking Models:
- Gemini 3 Pro (Thinking): Google’s most advanced reasoning model, over 50% improvement in solved benchmark tasks vs Gemini 2.5 Pro, with 1M token context window
- Gemini 2.5 Flash (Thinking): High-speed thinking model from Google
Non-Thinking Models:
- Gemini 2.5 Flash: Google’s best high-speed model without thinking
- Gemini 2.0 Flash: Excellent price-performance ratio
Google’s … well, Google. They make both closed source Gemini models that compete at the frontier, and open weight models (Gemma) anyone can run. Gemini 3 Pro represents a major leap in multimodal understanding across text, images, audio, and video.
xAI Models
- Grok 4 (Thinking): xAI’s flagship thinking model, trained with reinforcement learning to use tools like code interpreter and web browsing. First model to score 50% on Humanity’s Last Exam.
- Grok 4 Fast: Cost-efficient version with similar performance to Grok 4 but 40% fewer thinking tokens and a 2 million token context window
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. Grok 4 achieved groundbreaking scores on ARC-AGI-2, nearly doubling the previous commercial state-of-the-art.
Meta Models
(Meta owns Facebook, Instagram, Threads, WhatsApp, Oculus, etc. It’s just like Google is technically part of Alphabet alongside Waymo etc.)
- Llama 4 Maverick: Best pound for pound contender - fast, reasonably priced, comparable to DeepSeek V3
- Llama 3.3 70b: Pound-for-pound excellence from the previous generation (now outdated but still available)
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).
Deepseek Models
- Deepseek Chat V3.1 (Thinking): Latest thinking model from DeepSeek with excellent reasoning capabilities. We use more expensive providers that do not train on output, instead of DeepSeek itself (which does train on your output).
- Deepseek Chat V3.1: Great and inexpensive non-thinking model from DeepSeek. We use more expensive providers that do not train on outputs, for privacy reasons.
DeepSeek is an amazing Chinese AI lab producing excellent models at remarkably low costs. We prioritize your privacy by using providers that don’t train on your conversations.
Qwen Models
- Qwen 3 235b (Thinking): Alibaba’s largest thinking model with 235 billion parameters
- Qwen 3 235b Instruct: Non-thinking variant of the 235b model
Alibaba’s Qwen models provide excellent performance across various sizes, with the Qwen 3 series representing their latest advancements in large language models.
MiniMax Models
- MiniMax M2: MiniMax’s latest flagship model, a 230B parameter MoE with only 10B active parameters. Built for elite performance in coding and agentic tasks. Highest-scoring open-weight model globally, following closely behind GPT-5 (high) and Grok 4. At only 8% of the price of Claude Sonnet and twice the speed.
MiniMax is a Chinese AI company backed by Alibaba and Tencent, making exceptional models. M2 achieved an unprecedented score for an open model on intelligence benchmarks, surpassing Google DeepMind’s Gemini 2.5 Pro.
Z-AI Models (Zhipu AI)
- GLM 4.6 (Thinking): Z-AI’s flagship thinking model with 355B parameters (35B active), featuring a 200K token context window. Near parity with Claude Sonnet 4 on real-world coding tasks while using ~15% fewer tokens.
Z-AI (Zhipu AI) is a Chinese AI company making great value models with strong reasoning. GLM 4.6 is fully open-weight with MIT license, allowing enterprises to self-host and customize.
Moonshot AI Models
- Kimi K2 (Thinking): Moonshot’s flagship thinking model, one of the best open LLMs available with agentic capabilities that beat GPT-5 and Claude Sonnet 4.5 in certain tasks. Can automatically select 200-300 tools to complete tasks autonomously.
- Kimi K2 0905: Updated non-thinking variant with improved coding performance and 256K token context window, think ChatGPT default
Moonshot AI’s Kimi K2 is a 1 trillion parameter MoE model (32B active) backed by Alibaba. The K2 Thinking variant is currently the most powerful open source thinking model available.
Image, Audio & Video Models
We use a combination of models that update as we find new price/performance champions for different aspects of the site. These may include:
- Flux (Kontext, Krea etc.)
- SeeDance/SeeDream
- Veo/Imagen
- GPT-Image
- ElevenLabs models
- Current state of the art speech-to-text and text-to-speech models
These get frequently updated, like our blends.
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.
Retired Models
These models have been retired from our primary offerings. They may still be available upon request but have been superseded by newer versions.
OpenAI: GPT-5, GPT-5 Low/High, GPT-5 Chat, GPT-OSS-20b, o3, o4-mini, GPT-4.1 Mini, GPT-4.1 Nano, GPT-4o mini, GPT-3.5 Turbo
Anthropic: Claude Opus 4 / Sonnet 4, Claude 3.5 Haiku
Google: Gemini 2.5 Pro (Thinking), Gemini 2.0 Flash Lite
Meta: Llama 4 Scout, Llama 3.1 8b, Llama 3.1 405b
xAI: Grok 3 Mini (Thinking)
Qwen: Qwen 3 32b (Thinking), Qwen 3 30b MoE
Deepseek: Deepseek Chat V3, Deepseek R1
Z-AI: GLM 4.5 / GLM 4.5 Air
MiniMax: MiniMax-01
Mistral AI: Mistral Small V3.2
Cohere: Command-A
