What every major AI model actually costs, per million tokens.

Highest-capability long-running agents and demanding knowledge work

Fast, low-cost text and vision workloads

Retired except on Bedrock/Google Cloud

High-volume, latency-sensitive agents and assistants

Complex analysis, coding, and long-running agent tasks

Complex agentic coding and enterprise knowledge work

Balanced production coding, agents, and enterprise workflows

Introductory pricing, ends 2026-08-31

Balanced production coding, agents, and enterprise workflows

Fast, economical reasoning and straightforward agent tasks

Advanced reasoning, STEM, coding, and agentic workflows

Fast multimodal processing and high-volume agent workflows

Text/image/video input; audio input billed at $1.00

Low-cost, high-throughput classification and extraction

Text/image/video input; audio input billed at $0.30

Complex reasoning, coding, and long multimodal documents

Price rises with prompt length; shown value is the lower tier -- see source for the upper tier ($2.50 / $0.25 / $15.00)

Frontier multimodal reasoning and complex agentic tasks

Price rises with prompt length; shown value is the lower tier -- see source for the upper tier ($4.00 / $0.40 / $18.00)

Agentic coding and sustained multimodal workflows

Lowest-cost high-throughput execution in the Gemini 3.5 family

Balanced agentic and multimodal tasks with efficient token use

Cost-efficient multimodal reasoning and agent tasks

Kimi notes that web_search is currently being updated and does not recommend it in the near term.

Coding agents, visual understanding, and flexible thinking workloads

Kimi notes that web_search is currently being updated and does not recommend it in the near term.

Long-context programming, deep reasoning, and coding agents

High-speed long-context programming and coding agents

Same model as Kimi K2.7 Code, served at approximately 180 tokens/s and up to 260 tokens/s for short contexts.

Long-horizon coding, deep reasoning, and end-to-end knowledge work

Kimi notes that web_search is currently being updated and does not recommend it in the near term.

Long-document text generation and analysis

Image understanding with long-document context

Medium-length documents and general text generation

Image understanding with medium-length context

Short-context general text generation

Short-context image understanding and text generation

Low-latency code completion, fill-in-the-middle, and generation

Premier low-latency coding model.

Autonomous software engineering and agentic coding

Open-weights agentic coding model.

Lightweight multimodal coding agents

Labs model for lightweight coding agents.

Domain-specific, transparent, and multilingual reasoning

Premier thinking model.

Cost-efficient multilingual reasoning and multimodal analysis

Premier lightweight thinking model.

Higher-quality edge inference and compact agent workloads

Open lightweight edge model.

Small edge deployments and lightweight agents

Open lightweight edge model.

Balanced edge inference and lightweight agents

Open lightweight edge model.

General-purpose multilingual and multimodal enterprise workloads

Open-weight flagship multimodal and multilingual model.

Enterprise reasoning, coding, agents, and multimodal work

Open model. State-of-the-art performance with simplified enterprise deployment.

Low-cost lightweight coding workloads

Open lightweight model trained specifically for code tasks.

Cost-efficient multilingual, multimodal, and agentic workloads

Open Apache 2.0 model.

Higher-quality open-weight general text generation

Open sparse Mixture-of-Experts model.

Lightweight general text generation with open weights

Open sparse Mixture-of-Experts model.

Testing the latest Instant model experience used in ChatGPT

Latest Instant model used in ChatGPT; its underlying snapshot is regularly updated. OpenAI recommends GPT-5.6 for production API usage.

Long-running software engineering and agentic coding

Specialized Codex model for agentic software engineering.

General complex reasoning, coding, and professional workflows

Affordable model for coding and professional work. Prompts over 272K input tokens use long-context pricing for the full request: $5 input, $0.5 cached input, $22.5 output per 1M. Eligible regional-processing endpoints add 10%.

Cost-efficient production agents and routine reasoning tasks

Strong mini model for coding, computer use, and subagents. Eligible regional-processing endpoints add 10%.

High-volume classification, extraction, and simple automation

Lowest-cost GPT-5.4-class model for simple high-volume tasks. Eligible regional-processing endpoints add 10%.

Highest-quality GPT-5.4 reasoning for difficult professional work

Higher-compute GPT-5.4 model for maximum response quality. Prompts over 272K input tokens use long-context pricing for the full request: $60 input, $270 output per 1M. Eligible regional-processing endpoints add 10%.

Complex production reasoning, coding, and tool-driven workflows

Frontier coding and professional-work model. Prompts over 272K input tokens use long-context pricing for the full request: $10 input, $1 cached input, $45 output per 1M. Eligible regional-processing endpoints add 10%.

Maximum-quality GPT-5.5 reasoning and complex professional tasks

Higher-compute GPT-5.5 model for maximum response quality. Prompts over 272K input tokens use long-context pricing for the full request: $60 input, $270 output per 1M. Eligible regional-processing endpoints add 10%.

Cost-sensitive, high-volume production workloads

GPT-5.6 model for cost-sensitive, high-volume workloads. Prompts over 272K input tokens use long-context pricing for the full request: $2 input, $0.2 cached input, $2.5 cache write, $9 output per 1M. Eligible regional-processing endpoints add 10%.

Frontier complex reasoning, coding, and professional work

Frontier GPT-5.6 model for complex professional work. Prompts over 272K input tokens use long-context pricing for the full request: $10 input, $1 cached input, $12.5 cache write, $45 output per 1M. Eligible regional-processing endpoints add 10%.

Strong intelligence with balanced cost and latency

GPT-5.6 model balancing intelligence and cost. Prompts over 272K input tokens use long-context pricing for the full request: $5 input, $0.5 cached input, $6.25 cache write, $22.5 output per 1M. Eligible regional-processing endpoints add 10%.

Low-cost multimodal understanding and fast agent workflows

International list price shown for requests up to 256K tokens. Above 256K through 1M: $1 input and $4 output. Explicit cache hits cost 10%; implicit hits cost 20%.

Ultra-low-cost, high-throughput reasoning and text generation

International list price shown for requests up to 32K tokens. 32K-256K: $0.10 input/$0.40 output; 256K-1M: $0.20 input/$0.80 output. Explicit cache hits cost 10%; implicit hits cost 20%.

Highest-capability Qwen reasoning and text generation

Global list price through 1M tokens. Explicit cache hits cost 10% and implicit cache hits 20% of input list price. Time-limited night/day discounts may apply.

Balanced multimodal reasoning, long video, and tool use

Global list price shown for requests up to 256K tokens. Above 256K through 1M: $0.826 input and $3.301 output. Explicit cache hits cost 10%; implicit hits cost 20%. Time-limited night/day discounts may apply.

Fast general-purpose responses and tool-driven workflows

Short-context prices shown. At prompts of 200K tokens or more, all tokens cost $2.5 input, $0.4 cached input, and $5 output per 1M. Batch API discount: 20%. Priority processing costs 2x standard rates.

Deep reasoning, analysis, and tool-driven workflows

Short-context prices shown. At prompts of 200K tokens or more, all tokens cost $2.5 input, $0.4 cached input, and $5 output per 1M. Batch API discount: 20%. Priority processing costs 2x standard rates.

Parallel multi-agent deep research and complex investigations

Short-context prices shown. At prompts of 200K tokens or more, all tokens cost $2.5 input, $0.4 cached input, and $5 output per 1M. Batch API discount: 20%. Priority processing costs 2x standard rates.

General reasoning and tool-driven production workflows

Short-context prices shown. At prompts of 200K tokens or more, all tokens cost $2.5 input, $0.4 cached input, and $5 output per 1M. Batch API discount: 20%. Priority processing costs 2x standard rates.

Agentic software engineering and complex coding workflows

Short-context prices shown. At prompts of 200K tokens or more, all tokens cost $4 input, $0.6 cached input, and $12 output per 1M. Priority processing costs 2x standard rates.

Fast agentic coding, debugging, web development, and MCP

Short-context prices shown. At prompts of 200K tokens or more, all tokens cost $2 input, $0.4 cached input, and $4 output per 1M. Priority processing costs 2x standard rates.