Language model timeline

A new multimodal model for complex coding, knowledge work, and cyber defense, initially available to invited teams.

Improves coding, computer use, and professional work, with task delegation available within a request.

The second Claude 5.5 model supports coding, document processing, and office tasks.

Opus targets complex knowledge work, coding, and judgment-intensive tasks.

Two GPT-6 models address different speed and cost requirements.

MiMo combines multimodal understanding with large-scale reinforcement learning.

Grok improves coding, self-verification, and context management.

A flagship preview targets long-running engineering, knowledge work, and research.

Text, images, audio, and video share one multimodal model.

Kimi Code gains a stronger coding and agent preview.

A new architecture adds native vision and more efficient agent reasoning.

Flash improves sustained coding and professional reasoning.

Muse improves long-running agents, coding, and continued collaboration.

Improves engineering projects, multi-tool orchestration, chart reasoning, and document parsing.

Fable and Mythos improve coding, knowledge work, and research.

Tencent opens a flagship preview for long-context and complex work.

Native vision and hybrid linear/sparse attention support coding, browser, and GUI tasks.

An efficient multimodal model supports coding and high-volume agent use.

GLM strengthens complex coding and extended task execution.

V4 Pro reaches general availability with stronger agents and Responses API support.

Flash improves multi-step coding, knowledge work, and web development.

Grok improves sustained work across agents, interactive apps, and vision.

Qwen opens 2.4T-A95B weights, followed by a 27B model.

A native vision-language flagship supports a one-million-token context.

Opus upgrades engineering, knowledge work, and computer use.

Flash improves reasoning efficiency alongside a lighter Flash-Lite model.

Grok advances coding, agent tasks, and professional knowledge work.

Kimi combines native vision, long context, and complex coding.

Sol, Terra, and Luna address different knowledge-work and coding needs.

Muse improves coding, computer use, and multimodal understanding.

Tencent improves long-running agents and product integration.

Sonnet expands planning, browser use, terminal work, and sustained execution.

GLM extends long-running task and engineering capabilities.

Kimi improves long-running coding, tool calls, and efficient thinking.

Sparse attention adds native image and video understanding.

Step improves efficient reasoning and practical agent execution.

Opus improves tool use, complex judgments, and collaboration.

A public preview targets sustained coding and remote agent tasks.

Qwen improves agentic coding, office work, and long-running execution.

Google I/O introduces a Flash update focused on speed and agent tasks.

GPT improves tasks spanning multiple tools and applications.

MiMo upgrades complex tasks, multimodal work, and inference efficiency.

A flagship preview strengthens coding and complex task execution.

Opus improves difficult engineering and visual understanding.

GLM improves long-running engineering, planning, and execution.

Gemma advances open-model reasoning and agent capabilities.

MiMo expands flagship reasoning and multimodal understanding.

Software engineering and professional work gain stronger agent collaboration.

One model combines reasoning, vision, and coding with adjustable thinking.

Hybrid experts and long context support multi-agent work.

Reasoning, coding, and native computer use target professional work.

Gemini strengthens reasoning for scientific, engineering, and complex tasks.

Sonnet improves coding, visual interaction, planning, and knowledge work.

A smaller model previews rapid, interactive coding.

Reinforcement learning targets practical coding, search, office work, and tools.

Opus expands complex codebase analysis and long-running agent tasks.

Coding and general reasoning combine in longer engineering workflows.

Sparse mixture-of-experts modeling balances speed with agent capabilities.

A lightweight model extends the GLM-4.7 family.

MiniMax improves programming across languages and practical office tasks.

GLM improves coding and long-context task execution.

Gemini 3 reasoning reaches a faster, more efficient Flash tier.

Xiaomi opens an efficient model for reasoning, coding, and agent tasks.

NVIDIA announces Nemotron 3 and releases Nano first.

GPT advances professional knowledge work, coding, and long-context tasks.

Mistral expands its open-weight multimodal model family.

Configurable extended thinking supports reasoning, coding, and agent workflows.

DeepSeek advances efficient attention, reasoning, and agentic tool use.

Opus improves coding, computer use, and long-running work.

Instant and Thinking improve instructions, style, and adaptive reasoning.

Kimi extends the K2 family with deeper reasoning and tool-assisted work.

MiniMax opens a model designed around coding and agent workflows.

A smaller Claude model improves coding and computer use at lower latency.

GLM improves code, reasoning, and tool calls.

Sonnet advances coding, computer use, and sustained agentic tasks.

Hybrid attention and highly sparse activation target efficient long-context processing.

Thinking and non-thinking modes are unified with stronger agent capabilities.

Opus improves coding, practical reasoning, and agent performance.

An open mixture-of-experts model targets agentic coding and long tasks.

xAI advances reasoning and tool-assisted problem solving.

A major R1 update improves reasoning, coding, structured output, and tool calls.

An enterprise model balances multimodal capability, coding, and deployment cost.

An API model family improves coding, instructions, and long-context processing.

Tencent introduces a deep-reasoning model trained with reinforcement learning.

The open-weight family expands image understanding and long-context processing.

A research preview explores further gains from scaling pretraining.

xAI describes reasoning capabilities and the DeepSearch experience.

Efficient reasoning targets mathematics, science, and programming.

Synthetic data and training improvements strengthen small-model reasoning.

A revised 70B instruction model improves multilingual dialogue and deployment efficiency.

The full o1 model launches with ChatGPT Pro and adds image understanding.

Amazon introduces text and multimodal models across several cost tiers.

Vision models and small on-device text models join the Llama family.

General, coding, and math models improve instruction following and structured output.

A 123B model expands coding, multilingual, and long-context capabilities.

A lower-cost text and vision model expands lightweight applications.

A redesigned architecture introduces efficient 9B and 27B open-weight models.

A range of model sizes improves multilingual, coding, math, and long-context performance.

Microsoft introduces small language models suited to local deployment.

A sparse mixture-of-experts model expands efficient open-weight inference.

A developer preview expands context to 128K and advances vision and tool use.

xAI announces Grok with early access and information from X.

Improved coding, reasoning, and long-document processing arrive with a public chat beta.

Google explores reasoning and language capabilities at larger scale.