Find the AI model that fits the job.

Sustained agent work, writing, and high-end conversation

• chat
• reasoning
• coding
• research
• sites

• 1M-token context
• Adaptive thinking (always on)
• 128K max output tokens

Low-latency chat and lightweight coding

• speed
• chat
• coding
• sites

• Fastest comparative latency in the Claude lineup
• Extended thinking
• 64K max output tokens

Hard software, long agent sessions, and existing-site refactors

• coding
• reasoning
• chat
• research
• sites

• 1M-token context
• Adaptive thinking
• 128K max output tokens

Everyday coding, chat, and website updates with fast latency

• chat
• coding
• reasoning
• speed
• sites

• 1M-token context
• Adaptive thinking
• 128K max output tokens

Cheap, fast coding, chat, and light site edits

• speed
• coding
• chat
• sites

• Very low API price (reported)
• 1M-token context (reported)

Long-context research, reasoning, Workspace work, and site updates

• research
• reasoning
• chat
• coding
• sites

• 1M-token context (reported)
• Native Google Search grounding (reported)
• Workspace integration

High tokens-per-second chat and cheap inference

• speed
• chat

• Very high output speed (reported)
• 1M-token context (reported)
• Introductory Flash pricing

Sustained coding agents and large-repo website updates

• coding
• sites
• reasoning
• chat

• 1M-token context (Z.ai)
• Thinking effort levels for coding (Z.ai)
• MIT open-source license (Z.ai)

High-volume, latency-sensitive chat

• speed
• chat

• Lowest-cost GPT-5.6 tier
• Built for speed

Highest-capability GPT-5.6 tier for hard reasoning, coding, and site updates

• reasoning
• coding
• chat
• research
• sites

• Flagship GPT-5.6 tier
• Default family in ChatGPT (with Terra/Luna)
• API access

Daily ChatGPT-class work and website edits at mid-tier price

• chat
• coding
• reasoning
• speed
• sites

• Mid-tier GPT-5.6
• ChatGPT default family
• API access

Chat, coding, agentic work, and website updates with configurable reasoning

• chat
• reasoning
• coding
• research
• speed
• sites

• 500K-token context window
• Configurable reasoning
• Agentic tool calling

Spoken conversations, text-to-speech, and transcription

• voice
• chat

• Realtime conversations
• Text-to-speech
• Speech-to-text (batch and streaming)

Coding and long-context chat with published weights

• coding
• chat
• reasoning
• sites

• Open weights (reported)
• ~1M-token context (reported)
• API + self-host

Self-hosted or hosted chat, coding, and large-site edits

• chat
• coding
• sites
• reasoning

• Natively multimodal Llama 4 (Meta)
• 10M-token context window (Meta)
• Open weights for self-hosting

General chat, reasoning, and capable coding including site work

• chat
• reasoning
• coding
• sites

• Sparse MoE: 41B active / 675B total parameters (Mistral)
• Apache 2.0 open weights (Mistral)
• Multimodal (Mistral)

Agentic coding and website updates at a compact size

• coding
• sites
• reasoning
• chat
• speed

• 256K context window (Mistral)
• Configurable reasoning effort (Mistral)
• Unifies Magistral, Pixtral, and Devstral lines (Mistral)

Long-context coding, website updates, and professional work

• coding
• sites
• chat
• reasoning
• research

• 2.4 trillion parameters (Alibaba)
• Up to 1 million token context (Alibaba)
• Multimodal: text and visual intelligence (Alibaba)