LLM API pricing, tracked from launch.

OpenAI's reasoning model designed for complex problem-solving with extended chain-of-thought processing and a 200k token context window.

OpenAI's frontier model from the GPT-5.4 series with a 1M+ token context window, released in March 2026 as a unified system combining reasoning, coding, and general capabilities.

Highest-accuracy reasoning variant of GPT-5.5.

GPT-5.2 Pro is OpenAI's frontier-grade model with adaptive reasoning and a 400k token context window.

OpenAI's general-purpose reasoning model with a 400k token context window, released in August 2025 before its GPT-5.5 successor.

Premium high-compute reasoning variant launched alongside the 80% o3 price cut — 87% cheaper than the o1-pro it replaced.

o3 Pro is OpenAI's high-compute reasoning variant in the o-series, designed for problems requiring extensive chain-of-thought processing.

Anthropic's most powerful model — a new Mythos-class tier above Opus, aimed at the hardest reasoning and agentic work.

o3 Deep Research is OpenAI's specialized model designed for in-depth research and complex analysis tasks with extended reasoning capabilities.

OpenAI's flagship reasoning model with a 1M token context window, built for complex reasoning, coding, and multi-step problem-solving tasks.

Fugu Ultra is Sakana's high-performance multi-agent orchestration system that routes requests across specialized models rather than using a single monolithic architecture.

GPT-4 Turbo Preview is OpenAI's intermediate-tier large language model with a 128K token context window, released before the multimodal GPT-4o family.

OpenAI's frontier model for complex professional workloads. 1M token context, text + image input, native computer use.

Claude Opus 5 is Anthropic's flagship Opus-tier model, designed for complex reasoning, software development, and extended agentic tasks, with a 1M token context window.

Introduced the 67% Opus price cut — from $15/$75 (Opus 4/4.1) to $5/$25 — which held through all subsequent Opus releases.

First Opus with a 1M context window at standard pricing. Same $5/$25 as Opus 4.5.

Previous-generation Opus. Highly autonomous; strong on agentic, vision and memory tasks.

Most capable Opus-tier model: highly autonomous, strong on long-horizon agentic work and knowledge tasks. 1M context at standard pricing.

Kimi K3 is a multimodal reasoning model from Moonshot AI with a 1M+ token context window designed for complex problem-solving and long-context tasks.

A multimodal large language model from OpenAI that processes text and images with a 128K token context window.

Sonar Pro is a large language model from Perplexity with a 200k token context window that integrates web search capabilities.

OpenAI's general-purpose model from the GPT-5.4 series with a 1.05M token context window, released in early 2026 as the standard variant bridging lightweight and professional-grade tiers.

Previous Sonnet generation. Same $3 / $15 pricing as 4.6 — Sonnet pricing has held flat.

Anthropic's best balance of speed and intelligence. 1M context window. Sonnet pricing has held flat at $3/$15 across all 4.x releases.

GPT-5.2 is OpenAI's general-purpose reasoning model with a 400k token context window, released in December 2025 as the successor to GPT-5.1.

A coding-specialized model from OpenAI's GPT-5.2 generation with a 400k token context window, released in December 2025 as a variant optimized for software engineering tasks.

A coding-focused model from OpenAI's GPT-5.3 generation with a 400k token context window, released after GPT-5.3 Chat as a specialized variant for software engineering tasks.

GPT-5.3 Chat is OpenAI's conversational model from the GPT-5.3 generation with a 128k token context window.

Amazon Nova Premier is the most capable of Amazon’s multimodal models for complex reasoning tasks and for use as the best teacher for distilling custom models.

Previous Gemini Pro generation with the same context-tiered pricing model. Superseded by Gemini 3.1 Pro.

Google's frontier model. Context-tiered pricing: $2/$12 up to 200K tokens, $4/$18 beyond. Figures shown are the ≤200K tier.

Claude Sonnet 5 is Anthropic's mid-tier model balancing speed and capability, with a 1M token context window.

Cohere's enterprise RAG flagship, marketed as Command R+. The August 2024 refresh cut pricing from $3/$15 to $2.50/$10.

A multimodal large language model from OpenAI that processes text and images, released in August 2024 with a 128K token context window.

Cohere's August 2024 update to Command R+, a large language model with a 128K token context window optimized for production workloads.

Inflection 3 Pi is a conversational model that powers the Pi chatbot, designed for dialogue with built-in knowledge of recent events.

Inflection 3 Productivity is a compact instruction-following model optimized for structured outputs and rule-based tasks with access to recent information.

A multimodal large language model that processes text and images, with native support for structured outputs via JSON schema validation.

GPT-4o with integrated web search capability, allowing real-time retrieval and citation of current information from the internet during inference.

Command A is a 111 billion parameter open-weights model from Cohere with a 256K token context window, designed for agentic, multilingual, and coding tasks.

GPT-5 Chat is OpenAI's general-purpose conversational model with a 128k token context window, positioned as the baseline chat variant within the GPT-5 family.

GPT-5 Codex is OpenAI's code-focused model with a 400k token context window, built for software development and code generation tasks.

GPT-5.1-Codex is OpenAI's specialized model for software engineering tasks, built on the GPT-5.1 architecture with a 400k token context window.

GPT-5.1 is OpenAI's general-purpose reasoning model with a 400k token context window, released before the December 2025 GPT-5.2 successor.

GPT-5.1-Codex-Max is OpenAI's expanded-capacity variant of GPT-5.1-Codex, optimized for software engineering tasks with a 400k token context window.

GPT Audio is OpenAI's multimodal model that processes text, images, and audio inputs within a 128k token context window.

First Gemini with native thinking/reasoning. Context-tiered: $1.25/$10 ≤200K, $2.50/$15 beyond.

Launched at commodity pricing ($1.25/$10) so low TechCrunch said it may spark a price war. Superseded by GPT-5.5.

Latest Flash model — 3× pricier than Gemini 3 Flash ($0.50/$3), signaling the end of ultra-cheap Flash pricing.

Aion-1.0 is a large language model from AionLabs with a 131k-token context window, released before the company's shift toward multi-model systems.

A research-focused model from Perplexity that autonomously searches, retrieves, and synthesizes information across complex topics with a 128k token context window.

A reasoning model based on DeepSeek R1 that uses chain-of-thought processing to work through complex problems step-by-step, with access to Perplexity's search integration.

Jamba Large 1.7 is a hybrid SSM-Transformer model from AI21 with a 256K token context window, designed for efficient processing of long documents and complex tasks.

OpenAI's specialized model for in-depth research and complex analysis tasks, with extended reasoning capabilities and a 200k token context window.

Workhorse text model with 1M context. Positioned between the budget GPT-5 and premium GPT-5.5.

OpenAI's frontier reasoning model. Received an 80% price cut in June 2025 alongside the o3-pro launch.

Gemini 3.6 Flash is Google's high-efficiency text model optimized for coding, agentic workflows, and application development with a 1M token context window.

A midrange model in OpenAI's GPT-5.6 series that balances capability and efficiency with a 1.05M token context window.

Alibaba's latest flagship Qwen model with 1M context, announced at Alibaba Cloud Summit.

Consolidated model folding Medium 3.1, Magistral reasoning, and Devstral 2 into one set of weights with a per-request reasoning toggle.

Qwen3.6 Max Preview is Alibaba's preview flagship model from the Qwen3.6 generation with support for text and image inputs and a 262K token context window.

Grok 4.5 is xAI's frontier-tier model designed for advanced reasoning across coding, knowledge work, and STEM domains.

A multi-model system from AionLabs that combines specialized models for collaborative generation, designed around roleplaying and storytelling tasks.

Mistral's instruction-tuned mixture-of-experts model with 39B active parameters selected from 141B total, designed for efficient inference across diverse tasks.

Mistral's flagship model with 131k token context window, designed for reasoning, code generation, JSON formatting, and conversational tasks.

Writer's most advanced model designed for building and scaling AI agents in enterprise environments, with a context window supporting up to 1 million tokens.

This is a series of models designed to replicate the prose quality of the Claude 3 models, specifically Sonnet(https://openrouter.ai/anthropic/claude-3.5-sonnet) and Opus(https://openrouter.ai/anthropic/claude-3-opus). The model is fine-tuned on top of [Qwen2.5 72B](https://openrouter.ai/qwen/qwen-2.5-72b-instruct).

Fastest and most cost-effective Claude model. Pricing held flat since launch — no newer Haiku has been released.

A lightweight variant of GPT-5.4 with a 400k token context window, released in March 2026 as a smaller and faster alternative within the GPT-5.4 series.

o3 Mini High is OpenAI's cost-effective reasoning model with extended chain-of-thought processing, multimodal input support, and a 200k token context window.

OpenAI's o4-mini successor to o3-mini High, designed for cost-effective reasoning with extended chain-of-thought processing and multimodal input support.

Cost-effective reasoning model with multimodal support. Replaced o3-mini at the same price point.

Meta's first paid model API — a multimodal reasoning model built for agentic tasks, tool/computer use, coding, and multimodal understanding. Meta AI/Superintelligence Labs' answer to Anthropic/OpenAI's agentic-coding models.

Inkling is an open-weight mixture-of-experts model with 41B active parameters out of 975B total, supporting multimodal inputs and a 524K token context window.

GLM-5V Turbo is Z.ai's native multimodal model that processes image, video, and text inputs for agent-driven tasks with a 202K token context window.

GPT-3.5 Turbo with a 16,385-token context window, an expanded version of OpenAI's lightweight language model.

GLM 5 Turbo is Z.ai's optimized inference variant of the GLM 5 foundation model, designed for faster response times while maintaining broad capability across reasoning and long-context tasks.

Moonshot AI's latest open-weight frontier model with 300-agent swarm support.

Qwen3 Max Thinking is Alibaba's extended reasoning model with a 262K token context window, designed to perform deliberative reasoning across complex tasks.

Kimi K2.7 Code is a specialized language model from Moonshot AI with a 262K token context window optimized for software development tasks.

A routing system that analyzes requests and directs them to specialized LLMs from a managed library based on task requirements.

Qwen3 Coder Plus is Alibaba's mid-tier coding model with a 1M-token context window, positioned between the lightweight Coder Flash and larger general-purpose models in the Qwen3 lineup.

A multimodal model from Amazon designed to balance accuracy and speed across a broad range of tasks, with a 300K-token context window.

GLM 5.1 is Z.ai's foundation model with a 200K-token context window, positioned between the earlier GLM 5 and the newer reasoning-focused GLM 5.2.

GPT-5.6 Luna is OpenAI's lightweight model in the GPT-5.6 series, optimized for speed and efficiency across a wide range of tasks.

This is [Sao10K](/sao10k)'s experiment over [Euryale v2.2](/sao10k/l3.1-euryale-70b).

A 235B Mixture-of-Experts model from Alibaba that activates 22B parameters per forward pass and supports extended reasoning with a 262K token context window.

A 405-billion parameter reasoning model based on Llama 3.1 that can choose between direct answers and extended internal deliberation modes.

Relace Search is an agentic code search tool that uses parallel file and grep operations to explore codebases and locate relevant files.

Multimodal model with Agent Swarm technology. Superseded by K2.6.

Cost-effective Flash model. Superseded by Gemini 3.5 Flash which tripled the price.

A coding-focused language model from Kwaipilot with a 256K token context window.

A 235-billion-parameter multimodal vision-language model from Alibaba that processes text, images, and video with extended reasoning capabilities and a 131k-token context window.

GLM 5 is Z.ai's foundation model with a 198K-token context window for processing extended documents and conversations.

Google's lightweight text model designed for efficient subagent execution within multi-agent workflows, with a 1M token context window.

DeepSeek R1 is a 671B parameter reasoning model with 37B active parameters that shows performance comparable to OpenAI o1, featuring fully open reasoning tokens.

Kimi K2 0905 is a Mixture-of-Experts language model from Moonshot AI with a 262K token context window, released in September 2025.

Kimi K2 Thinking is a reasoning-focused model from Moonshot AI that uses extended chain-of-thought to work through complex problems before generating responses.

Nova 2 Lite is Amazon's lightweight multimodal model that handles text, images, and video input with a 1M token context window.

A multi-agent variant of xAI's Grok 4.20 with a 2 million token context window, released before Grok 4.3.

Predecessor to Grok 4.3 with 2M context. Later aligned to Grok 4.3 pricing.

xAI's current flagship — aggressively priced for a frontier model, with a 1M token context window.

First Flash model with thinking. Flat pricing (no context tiers). Superseded by Gemini 3 Flash.

GLM 5.2 is a large-scale reasoning model from Z.ai with a 1M-token context window for processing long documents and complex tasks.

A lightweight multimodal model from OpenAI that processes text, images, and audio inputs, positioned as a smaller variant of GPT Audio released in January 2026.

Qwen3.5 397B A17B is a 397-billion-parameter multimodal model from Alibaba combining linear attention with sparse mixture-of-experts routing.

Kimi K2 0711 is a general-purpose language model from Moonshot AI with a 131K token context window, released in July 2025.

MiniMax M1 is an open-weight reasoning model with a 1M-token context window and hybrid mixture-of-experts architecture designed for efficient inference at scale.

GLM 4.5 is Z.ai's foundation model with a 131K-token context window for processing documents and conversations.

A 55B-parameter mixture-of-experts model from NVIDIA with a 262K token context window, using a hybrid Transformer-Mamba architecture for reasoning and task orchestration.

DeepSeek's 671-billion parameter reasoning model with fully transparent chain-of-thought reasoning tokens and a 163k token context window.

Qwen3.5-122B-A10B is a 122-billion-parameter multimodal model from Alibaba combining linear attention with sparse mixture-of-experts routing.

GPT-3.5 Turbo Instruct is OpenAI's instruction-tuned variant of GPT-3.5 Turbo, designed for text completion tasks rather than chat interactions.

GPT-3.5 Turbo v0613 is an older snapshot of OpenAI's lightweight language model, optimized for chat and code tasks with a 4K token context window.

Mistral Medium 3 is an enterprise language model that handles complex reasoning tasks with a 131k-token context window.

GPT-5 Mini is OpenAI's lightweight general-purpose model with a 400k token context window, positioned as a smaller and faster variant within the GPT-5 family.

Mistral Medium 3.1 is a mid-sized language model from Mistral designed for enterprise applications, offering a balance between capability and computational efficiency with a 131k token context window.

GLM 4.6 is Z.ai's foundation model with a 202K-token context window for processing extended documents and conversations.

A lightweight variant of GPT-5.1-Codex optimized for software engineering tasks with reduced computational overhead.

A 123B dense transformer from Mistral designed for code generation and agentic workflows, with support for 256K token context.

Seed 1.6 is a general-purpose model released by the ByteDance Seed team. It incorporates multimodal capabilities and adaptive deep thinking with a 256K context window.

Seed-2.0-Lite is a multimodal model from ByteDance with a 262k token context window, designed for enterprise production workloads with emphasis on lower latency.

A 27-billion-parameter text model from Alibaba's Qwen3.6 generation with a 262K token context window.

xAI's coding-specialist model — cheaper than Grok 4.3 for code generation tasks.

Qwen3.6 Plus is Alibaba's mid-tier language model from the Qwen3.6 generation with a 1 million token context window.

Morph V3 Large is a code-specialized model designed for precise code transformations and edits at scale with a 262K token context window.

Qwen3 VL 235B A22B Instruct is a 235-billion-parameter multimodal vision-language model from Alibaba that processes text, images, and video within a 131k-token context window.

A 235B mixture-of-experts model from Alibaba's Qwen that activates 22B parameters per forward pass, with support for extended thinking mode for complex reasoning tasks.

GLM 4.5V is Z.ai's vision-language model that processes text, image, and video inputs for multimodal reasoning tasks.

Qwen3.5 Plus is Alibaba's mid-tier language model from April 2026 supporting a 1 million token context window.

GLM 4.7 is Z.ai's foundation model with a 202K-token context window for processing extended documents and conversations.

Aion-RP 1.0 is an 8-billion-parameter language model from AionLabs designed for roleplaying and narrative generation tasks.

Aion-2.0 is a multi-model system from AionLabs that builds on the reasoning and coding strengths of Aion-1.0 with broader capability across language tasks.

Fast, cost-effective model with 1M context for high-volume workloads.

A 30B parameter mixture-of-experts model from Alibaba designed for reasoning tasks with extended thinking capabilities and internal reasoning traces.

A 30B multimodal vision-language model from Alibaba that processes text, images, and video with extended reasoning capabilities and a 131k-token context window.

Qwen3.5 Plus is Alibaba's mid-tier language model from February 2026 with a 1 million token context window.

Qwen3.5-27B is a 27-billion-parameter text model from Alibaba with a 262K token context window.

OpenAI's lightweight language model optimized for chat and code tasks with a 16K token context window.

Google's lightweight text model designed for efficient execution in multi-agent and agentic workflows with a 1M token context window.

A vision-language model from Perceptron designed to process images and videos alongside natural language queries for visual understanding tasks.

Mistral's Apache-2.0 open-weight frontier model (675B total / 41B active MoE). 75% cheaper than Large 2 at $0.50/$1.50.

Alibaba's previous flagship. Received a 50% price cut in November 2025 as China's AI price war intensified.

Aion-3.0-Mini is a multi-model roleplaying and storytelling system from AionLabs built on the DeepSeek family, using collaborative generation from specialized models.

Aion-1.0-Mini is a smaller parameter variant of Aion-1.0 from AionLabs, sharing the same 131k-token context window as its full-size sibling.

An 8B multimodal vision-language model from Alibaba that processes text, images, and video with extended reasoning capabilities for structured problem-solving.

Qwen3.7 Plus is Alibaba's multimodal model supporting text, images, and video inputs with a 1 million token context window.

A multimodal mixture-of-experts model from Baidu with 424B total parameters and 47B active per token, trained on both text and image data.

A specialized code-patching model that applies AI-suggested edits directly into source files from multiple LLM providers.

Cogito v2.1 671B is a 671-billion parameter mixture-of-experts model trained with self-play reinforcement learning, offering a 128k token context window.

A lightweight variant of GPT-5.4 with a 400k token context window, positioned as a smaller, faster alternative within OpenAI's GPT-5.4 series.

A sparse mixture-of-experts model from Meituan with 48B active parameters designed to handle very long contexts up to 1M tokens.

Virtuoso Large is a 72B general-purpose language model from Arcee AI with a 131k token context window.

Morph's fastest model for applying code edits and transformations, optimized to process code modifications at high throughput.

MiniMax M2.1 is a 10-billion-parameter lightweight language model designed for coding, agentic workflows, and application development.

MiniMax M2-her is a dialogue-focused model designed for character-driven conversations and roleplay scenarios with a 65K token context window.

A code-focused language model from Kwaipilot with a 256K token context window.

MiniMax M3 is a multimodal foundation model that processes text, image, and video inputs to produce text output, with a 524K token context window.

StepFun's multimodal Mixture-of-Experts model that processes text, images, and video with a 196B parameter backbone while activating approximately 11B parameters per inference.

A lightweight text and vision model from Alibaba's Qwen family with a 1 million token context window.

DeepSeek V3 is a 685-billion-parameter mixture-of-experts model designed for multi-turn conversations and reasoning tasks.

MiniMax-01 is a multimodal model combining text generation and image understanding with 456 billion parameters and a 1 million token context window.

Qwen3-Next-80B-A3B-Instruct is an instruction-tuned 80B parameter model from Alibaba's Qwen3 series with a 262k token context window, designed for chat and task completion without extended reasoning traces.

A 106B-parameter mixture-of-experts model from Prime Intellect with 12B active parameters, post-trained with supervised fine-tuning and reinforcement learning on a 131k-token context window.

MiniMax M2 is a 10B activated parameter model with a 196K token context window, designed for coding and agentic workflows.

Nex-N2-Pro is a mixture-of-experts model from Nex AGI that processes text and image inputs with a 262K token context window.