Guide

Synthetic.new is a privacy-focused AI platform offering private access to multiple open-source LLMs through simple flat-rate subscriptions starting at $20/month for 125 requests per 5 hours or $60/month for 1250 requests. The platform provides access to 19+ always-on models including Llama 3 variants with up to 128K token context windows, specialized coding models, and task-specific LoRA adapters, with guaranteed privacy through no training on user data and automatic deletion within 14 days. Key features include OpenAI-compatible API for integration with tools like Roo, Cline, and Octofriend, web-based chat interface, on-demand model launching from Hugging Face repositories on cloud GPUs with separate per-minute billing, predictable pricing without per-token charges, and support for large context coding tasks. The platform prioritizes developer workflows and code generation with strong privacy guarantees and cost-effective access to powerful open-source models.

DeepInfra is a cloud inference platform providing fast and cost-effective access to a wide range of open-source AI models including Llama, Mistral, DeepSeek, and more through an OpenAI-compatible API. Key features include serverless inference with pay-per-token pricing, support for text generation, embeddings, and image models, low-latency responses powered by optimized GPU infrastructure, and easy integration with any OpenAI-compatible client or SDK.

Chutes.ai is a decentralized serverless AI compute platform built on Bittensor Subnet 64, enabling developers to deploy, run, and scale AI models without managing infrastructure. The platform processes nearly 160 billion tokens daily serving over 400,000 users with up to 90% lower costs than traditional providers through a distributed network of GPU miners compensated with TAO tokens. Key features include always-hot serverless compute with instant inference, model-agnostic support for LLMs, image, and audio models plus custom code, fully abstracted infrastructure handling provisioning and scaling automatically, standardized API access with OpenRouter integration, and open pay-per-use pricing. The roadmap includes long-running jobs, fine-tuning capabilities, AI agents, and Trusted Execution Environments for enhanced privacy, with a startup accelerator offering up to $20,000 in credits.

Moonshot AI is a Beijing-based AI platform offering the Kimi large language model API, with the flagship Kimi K2 being a state-of-the-art Mixture-of-Experts (MoE) model featuring 1 trillion total parameters and 32 billion activated parameters per query. Key features include an exceptional 256,000-token context window (the longest available for processing extended documents and conversations), strong coding and STEM performance competitive with GPT-4.1, native tool calling and function integration for agentic workflows, and stable large-scale training using the novel MuonClip optimizer on 15.5 trillion tokens. The platform provides OpenAI-compatible API access through the Kimi Open Platform with variants including Kimi-K2-Base (for fine-tuning) and Kimi-K2-Instruct (optimized for chat and autonomous tasks), supporting advanced multi-turn interactions, reasoning, research, and software development applications.

Gemini 2.5 Flash Image, a.k.a. "Nano Banana," is now generally available. It is a state of the art image generation model with contextual understanding. It is capable of image generation,...

Specialized Gemini 2.5 model for browser-control agents that automate UI tasks

Multimodal reasoning model for visual analysis, planning, and tool use

Qwen3-VL-30B-A3B-Thinking is a multimodal model that unifies strong text generation with visual understanding for images and videos. Its Thinking variant enhances reasoning in STEM, math, and complex tasks. It excels...

Qwen3-VL-30B-A3B-Instruct is a multimodal model that unifies strong text generation with visual understanding for images and videos. Its Instruct variant optimizes instruction-following for general multimodal tasks. It excels in perception...

GPT-5 Pro is OpenAI’s most advanced model, offering major improvements in reasoning, code quality, and user experience. It is optimized for complex tasks that require step-by-step reasoning, instruction following, and...

GPT-5 Pro is OpenAI’s most advanced model, offering major improvements in reasoning, code quality, and user experience. It is optimized for complex tasks that require step-by-step reasoning, instruction following, and...

Groq is the world's fastest AI inference platform powered by the proprietary LPU™ (Language Processing Unit) Inference Engine, purpose-built hardware designed specifically for running large language models at exceptional speed and low cost. The LPU architecture delivers 300-500 tokens per second with up to 18x faster processing than traditional GPUs through tensor streaming technology optimized for sequential computation and low-latency inference. GroqCloud provides API access to leading open-source models (Llama, Mixtral, Gemma) with Tokens-as-a-Service pricing, enabling developers to build production-ready AI applications with ultra-low latency and high throughput. Key features include deterministic performance, reduced memory bottlenecks, energy-efficient processing, real-time inference capabilities, and scalable cloud deployment with straightforward API integration.

Higher-accuracy GPT-5 tier for tough analysis, coding reviews, and planning

Higher-accuracy GPT-5 tier for tough analysis, coding reviews, and planning

Image model for prompt-driven generation, editing, and visual design workflows

GLM 4.6 from Fireworks AI - text input, 198,000 token context

Compared with GLM-4.5, this generation brings several key improvements: Longer context window: The context window has been expanded from 128K to 200K tokens, enabling the model to handle more complex...

Compared with GLM-4.5, this generation brings several key improvements: Longer context window: The context window has been expanded from 128K to 200K tokens, enabling the model to handle more complex agentic tasks. Superior coding performance: The model achieves higher scores on code benchmarks and demonstrates better real-world performance in applications such as Claude Code、Cline、Roo Code and Kilo Code, including improvements in generating visually polished front-end pages. Advanced reasoning: GLM-4.6 shows a clear improvement in reasoning performance and supports tool use during inference, leading to stronger overall capability. More capable agents: GLM-4.6 exhibits stronger performance in tool using and search-based agents, and integrates more effectively within agent frameworks. Refined writing: Better aligns with human preferences in style and readability, and performs more naturally in role-playing scenarios.

Gemini 2.5 Flash TTS from Google - text input, 32,768 token context

GLM 4.6 from Synthetic - text input, 200,000 token context

Flagship GLM model for hybrid reasoning, coding, and agentic engineering

GLM-4.6V from DeepInfra - text, image input, 204,800 token context

Late GLM-4 workhorse for coding agents, reasoning, and structured tasks

Late GLM-4 workhorse for coding agents, reasoning, and structured tasks

Late GLM-4 workhorse for coding agents, reasoning, and structured tasks

Flagship GLM model for hybrid reasoning, coding, and agentic engineering

Claude Sonnet 4.5 is Anthropic’s most advanced Sonnet model to date, optimized for real-world agents and coding workflows. It delivers state-of-the-art performance on coding benchmarks such as SWE-bench Verified, with...

Claude Sonnet 4.5 is Anthropic’s most advanced Sonnet model to date, optimized for real-world agents and coding workflows. It delivers state-of-the-art performance on coding benchmarks such as SWE-bench Verified, with...

DeepSeek-V3.2-Exp is an experimental large language model released by DeepSeek as an intermediate step between V3.1 and future architectures. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism...

Balanced Claude model for coding, analysis, agent workflows, and cost control

Balanced Claude model for coding, analysis, agent workflows, and cost control

Balanced Claude model for coding, analysis, agent workflows, and cost control

Balanced Claude model for coding, analysis, agent workflows, and cost control

Balanced Claude model for coding, analysis, agent workflows, and cost control

Balanced Claude model for coding, analysis, agent workflows, and cost control

Balanced Claude model for coding, analysis, agent workflows, and cost control

Balanced Claude model for coding, analysis, agent workflows, and cost control

Balanced Claude model for coding, analysis, agent workflows, and cost control

DeepSeek chat model for instruction following, coding, and analysis

Balanced Claude model for coding, analysis, agent workflows, and cost control

ChatGPT is OpenAI's conversational AI assistant now powered by the latest GPT-5 family of models released in 2025, featuring unified multimodal capabilities across text, images, and audio with advanced reasoning. The current lineup includes GPT-5 (flagship general-purpose model with real-time routing, faster responses, reduced hallucinations, and customizable personalities), o3 (advanced reasoning model for complex math, science, and programming with 88.9% AIME accuracy), o4-mini (cost-efficient reasoning at scale with 92.7% AIME accuracy), and GPT-5-Codex (specialized for dynamic coding tasks). Key features include autonomous tool use (web browsing, code execution, file operations), self-fact checking, multimodal unification, and variants from Nano to Pro for different use cases.

Claude is a next-generation AI assistant developed by Anthropic, featuring a family of state-of-the-art large language models trained to be safe, accurate, and helpful. The latest models include Claude Sonnet 4.5 (the world's best coding model with advanced agentic capabilities) and Claude Opus 4.1, both offering hybrid reasoning modes, 200K token context windows, and sophisticated vision capabilities. Key features include tool use for external API integration, code execution environments, multi-step workflow automation, files API, persistent memory management, and enterprise-grade security with deployment on AWS Bedrock and Google Cloud Vertex AI. Claude excels at complex reasoning, code generation, visual data interpretation, customer support, and building autonomous AI agents with natural, human-like conversations.

Google Gemini is Google DeepMind's advanced multimodal AI platform designed for the "agentic era," with the latest Gemini 2.5 family of models released in 2025 featuring breakthrough thinking and reasoning capabilities. The current lineup includes Gemini 2.5 Deep Think (most advanced reasoning model using parallel multi-agent reasoning for complex math and coding), Gemini 2.5 Pro (flagship thinking model with enhanced performance and 1M token context window), Gemini 2.5 Flash (fastest thinking model balancing speed and intelligence), and Gemini 2.5 Flash-Lite (optimized for cost-effective deployment). Key features include native tool use, multimodal understanding (text, images, audio, video), code execution integration, Google Search connectivity, reinforcement learning-enhanced reasoning, and the ability to think through responses before answering for improved accuracy.

OpenRouter is a unified API gateway that provides access to 400+ AI models from 50+ providers through a single OpenAI-compatible endpoint, eliminating vendor lock-in and simplifying multi-model integration. Key features include automatic fallbacks (seamlessly switching to backup models if primary fails), smart model routing with :floor (cheapest) and :nitro (fastest) options, standardized API normalization across all providers, and multi-modal support for text and images. OpenRouter uses pass-through pricing at exact provider rates plus a 5% platform fee (5.5% on credits), offers 13+ free models with daily limits, consolidated analytics dashboards, and enterprise-grade privacy controls with no code changes needed to switch between models like GPT-4, Claude, Gemini, Llama, and DeepSeek.

Mistral AI offers high-performance, cost-effective language models with a focus on efficiency and European AI development. Their models are known for strong reasoning and coding capabilities.

DeepSeek AI is a Chinese open-source AI company offering advanced large language models with the latest DeepSeek-V3.1 (released August 2025) combining both general-purpose and reasoning capabilities in a hybrid architecture. Key models include DeepSeek-V3.1 (flagship with 128K token context window, 43% improved multi-step reasoning, and dual thinking/non-thinking modes), DeepSeek-R1 (specialized reasoning model with chain-of-thought processing matching OpenAI o1 performance), and DeepSeek-VL2 (state-of-the-art vision-language model). Features include hybrid Mixture-of-Experts (MoE) architecture, extended context handling up to 1M tokens, enhanced tool calling for agentic workflows, 20-50% faster inference than previous versions, JSON output support, and fully open-source with MIT licensing. Access the platform at [deepseek.ai](https://deepseek.ai) with API documentation at [api-docs.deepseek.com](https://api-docs.deepseek.com).

Grok is SpaceXAI's flagship family of large language models designed to deliver truthful and insightful AI responses. The SpaceXAI API provides developers with access to powerful models including Grok-4 and Grok-2-1212 (with a 131K token context window), supporting multimodal capabilities like vision processing and image generation via Flux.1. Key features include function calling for API automation, compatibility with OpenAI/Anthropic SDKs, structured outputs, and enterprise-grade security with GDPR/HIPAA compliance. The platform offers Python and JavaScript SDKs for easy integration into applications ranging from conversational AI to complex workflow automation.

Perplexity AI is an AI-powered search engine that combines large language models with real-time web search to deliver accurate, cited answers with up-to-date information. Key features include Pro Search (access to GPT-5, Claude 4, Gemini 2.5 Pro, Grok 4, and proprietary Sonar models), Deep Research (autonomous multi-step research synthesizing hundreds of sources into comprehensive reports with PDF export), Comet Browser (free AI-powered Chromium browser with conversational web control and sidecar assistant for summaries and automation), and API access for enterprise integration. Pro subscribers enjoy unlimited Deep Research queries, advanced model selection, domain-specific searches, file uploads, and image generation/editing tools, while free users get limited access to core features with real-time citations. Access at [perplexity.ai](https://www.perplexity.ai) with API documentation available for Pro users.

Jina AI is a specialized AI platform providing best-in-class search infrastructure through embeddings, rerankers, web readers, and small language models for multilingual and multimodal data. Key offerings include jina-embeddings-v3 (570M parameter model supporting 89 languages with 8K token context and task-specific LoRA adapters for retrieval, clustering, and classification), jina-embeddings-v4 (3.8B parameter multimodal model unifying text and images), Reader API (converts any URL to LLM-ready Markdown using ReaderLM-v2), Reranker API (improves search result accuracy), and DeepSearch (comprehensive search agent combining web search, reading, and reasoning with OpenAI-compatible API). The platform features FlashAttention 2 optimization, 1024-dimensional embeddings, late-chunking for better snippet selection, and integration with popular frameworks, making it ideal for RAG systems and semantic search applications.

Fireworks.ai is a high-performance generative AI platform that provides the fastest inference for open-source LLMs and multimodal models through a developer-friendly API. The platform features a proprietary FireAttention engine delivering 50% faster speed and 250% higher throughput than standard engines, with support for popular models like LLaMA, Mixtral, DeepSeek, and Falcon. Key capabilities include serverless inference, advanced fine-tuning (LoRA, RLHF), function calling, batch processing, on-demand GPU access (NVIDIA H100/H200, AMD MI300X), and OpenAI-compatible APIs.

Uncensored and creative writing model based on Mistral Small 3.2 24B with good recall, prompt adherence, and intelligence.