AI Timeline 2024–2026
Major milestones in artificial intelligence — models, regulation, research, and products.
Qwen 3 Max — hybrid reasoning model
Alibaba releases Qwen 3 Max, a hybrid model capable of switching between fast and thinking modes. Strong Chinese language support at $1 per million input tokens.
Qwen 3 Max — hybrid reasoning model
Alibaba releases Qwen 3 Max, a hybrid model capable of switching between fast and thinking modes. Strong Chinese language support at $1 per million input tokens.
Anthropic publishes Constitutional AI v2 paper
Anthropic publishes a detailed technical report on Constitutional AI v2, describing how Claude's values are trained via a hierarchical set of self-generated principles.
Anthropic publishes Constitutional AI v2 paper
Anthropic publishes a detailed technical report on Constitutional AI v2, describing how Claude's values are trained via a hierarchical set of self-generated principles.
DeepSeek V3 — multimodal open-weights
DeepSeek releases V3 with multimodal capabilities and fully open weights, enabling the community to fine-tune and deploy vision-language models locally.
DeepSeek V3 — multimodal open-weights
DeepSeek releases V3 with multimodal capabilities and fully open weights, enabling the community to fine-tune and deploy vision-language models locally.
EU AI Act General-Purpose AI rules effective
GPAI provisions of the EU AI Act take effect, requiring major AI labs to publish training data summaries, conduct red-team testing, and register with the EU AI Office.
EU AI Act General-Purpose AI rules effective
GPAI provisions of the EU AI Act take effect, requiring major AI labs to publish training data summaries, conduct red-team testing, and register with the EU AI Office.
Kimi K2.5 — 2M context window
Moonshot AI releases Kimi K2.5 with a 2M token context window, strong code generation, and competitive pricing. Targets the long-document analysis market.
Kimi K2.5 — 2M context window
Moonshot AI releases Kimi K2.5 with a 2M token context window, strong code generation, and competitive pricing. Targets the long-document analysis market.
GPT-5 released — multimodal reasoning
OpenAI releases GPT-5 with native image, audio, and video understanding, superior instruction following, and dramatically reduced hallucination rates versus GPT-4o.
GPT-5 released — multimodal reasoning
OpenAI releases GPT-5 with native image, audio, and video understanding, superior instruction following, and dramatically reduced hallucination rates versus GPT-4o.
OpenAI reaches 500M weekly users
OpenAI announces ChatGPT has surpassed 500M weekly active users — up from 100M in January 2023 — driven by GPT-4o voice and the operator API ecosystem.
OpenAI reaches 500M weekly users
OpenAI announces ChatGPT has surpassed 500M weekly active users — up from 100M in January 2023 — driven by GPT-4o voice and the operator API ecosystem.
Claude Opus 4 launched
Anthropic releases the flagship Opus 4 with 200K context, improved reasoning, and tool use — achieving top benchmark scores across coding and analysis.
Claude Opus 4 launched
Anthropic releases the flagship Opus 4 with 200K context, improved reasoning, and tool use — achieving top benchmark scores across coding and analysis.
Gemini 2.5 Pro — thinking + 1M context
Google releases Gemini 2.5 Pro with expanded thinking mode, improved coding benchmarks, and the largest commercially available context window at 1M tokens.
Gemini 2.5 Pro — thinking + 1M context
Google releases Gemini 2.5 Pro with expanded thinking mode, improved coding benchmarks, and the largest commercially available context window at 1M tokens.
DeepSeek R2 released — $0.5/1M tokens
DeepSeek releases R2 with improved reasoning and coding at $0.5 per million input tokens, sustaining the price pressure it began with R1.
DeepSeek R2 released — $0.5/1M tokens
DeepSeek releases R2 with improved reasoning and coding at $0.5 per million input tokens, sustaining the price pressure it began with R1.
Mistral Large 3 — 128K, multilingual
Mistral AI releases Large 3 with a 128K context window, strong European language support, and partial open weights. Competes with GPT-4o at lower cost.
Mistral Large 3 — 128K, multilingual
Mistral AI releases Large 3 with a 128K context window, strong European language support, and partial open weights. Competes with GPT-4o at lower cost.
EU AI Act Prohibited Practices ban effective
The first enforceable provisions of the EU AI Act — banning biometric categorization and social scoring systems — become law across all EU member states.
EU AI Act Prohibited Practices ban effective
The first enforceable provisions of the EU AI Act — banning biometric categorization and social scoring systems — become law across all EU member states.
Grok 3 released — #1 Arena ELO briefly
xAI releases Grok 3 with 131K context and strong reasoning. Briefly achieves top Arena ELO before being overtaken by Claude Opus 4 and GPT-5.
Grok 3 released — #1 Arena ELO briefly
xAI releases Grok 3 with 131K context and strong reasoning. Briefly achieves top Arena ELO before being overtaken by Claude Opus 4 and GPT-5.
OpenAI DevDay 2025 — agents & memory
OpenAI's developer conference focuses on the Agents SDK, persistent memory, custom personas, and new context caching APIs enabling long-running autonomous agents.
OpenAI DevDay 2025 — agents & memory
OpenAI's developer conference focuses on the Agents SDK, persistent memory, custom personas, and new context caching APIs enabling long-running autonomous agents.
GPT-4.5 released
OpenAI releases GPT-4.5, an iteratively improved version of GPT-4o with better instruction following, reduced hallucination, and improved conversation quality.
GPT-4.5 released
OpenAI releases GPT-4.5, an iteratively improved version of GPT-4o with better instruction following, reduced hallucination, and improved conversation quality.
Llama 4 Scout — 10M context window
Meta releases Llama 4, featuring Scout with a 10 million token context window and Maverick with MoE architecture. Both are open-weight models beating GPT-4o.
Llama 4 Scout — 10M context window
Meta releases Llama 4, featuring Scout with a 10 million token context window and Maverick with MoE architecture. Both are open-weight models beating GPT-4o.
MCP (Model Context Protocol) reaches 1000+ servers
Anthropic's open standard for connecting AI assistants to tools crosses 1,000 community-built MCP servers in its directory, becoming the de facto agent tooling standard.
MCP (Model Context Protocol) reaches 1000+ servers
Anthropic's open standard for connecting AI assistants to tools crosses 1,000 community-built MCP servers in its directory, becoming the de facto agent tooling standard.
Anthropic raises $3.5B at $61.5B valuation
Anthropic closes a $3.5B funding round, bringing total raised to $10B+ and valuation to $61.5B. Amazon's total commitment reaches $8B.
Anthropic raises $3.5B at $61.5B valuation
Anthropic closes a $3.5B funding round, bringing total raised to $10B+ and valuation to $61.5B. Amazon's total commitment reaches $8B.
DeepSeek R1 shocks the market
Chinese lab DeepSeek releases R1, a reasoning model trained for $5.6M that rivals o1 on benchmarks. Causes Nvidia to lose $600B in market cap in a single day.
DeepSeek R1 shocks the market
Chinese lab DeepSeek releases R1, a reasoning model trained for $5.6M that rivals o1 on benchmarks. Causes Nvidia to lose $600B in market cap in a single day.
Google Gemini 2.0 Flash launched
Google releases Gemini 2.0 Flash with native multimodal output (text, images, audio, code) and 1M context, as a preview for its next-generation model family.
Google Gemini 2.0 Flash launched
Google releases Gemini 2.0 Flash with native multimodal output (text, images, audio, code) and 1M context, as a preview for its next-generation model family.
Microsoft Copilot Studio GA
Microsoft makes Copilot Studio generally available, allowing enterprises to build custom AI agents that integrate with Microsoft 365, Teams, and Power Platform.
Microsoft Copilot Studio GA
Microsoft makes Copilot Studio generally available, allowing enterprises to build custom AI agents that integrate with Microsoft 365, Teams, and Power Platform.
Claude Computer Use — controlling a PC
Anthropic introduces Computer Use in Claude 3.5 Sonnet (beta), allowing the model to move the mouse, type, and interact with any software like a human.
Claude Computer Use — controlling a PC
Anthropic introduces Computer Use in Claude 3.5 Sonnet (beta), allowing the model to move the mouse, type, and interact with any software like a human.
OpenAI o1 series — "thinking" models
OpenAI launches o1 and o1-mini, models that spend more compute at inference time to "think" through problems step by step. Achieves PhD-level reasoning on benchmarks.
OpenAI o1 series — "thinking" models
OpenAI launches o1 and o1-mini, models that spend more compute at inference time to "think" through problems step by step. Achieves PhD-level reasoning on benchmarks.
EU AI Act enters into force
The world's first comprehensive AI regulation officially enters into force, creating risk-based compliance requirements for AI systems deployed in the European Union.
EU AI Act enters into force
The world's first comprehensive AI regulation officially enters into force, creating risk-based compliance requirements for AI systems deployed in the European Union.
GPT-4o mini released — $0.15/1M tokens
OpenAI releases GPT-4o mini at $0.15 per million input tokens, replacing GPT-3.5 Turbo. Sets a new price floor for capable frontier models.
GPT-4o mini released — $0.15/1M tokens
OpenAI releases GPT-4o mini at $0.15 per million input tokens, replacing GPT-3.5 Turbo. Sets a new price floor for capable frontier models.
Claude 3.5 Sonnet sets new benchmark bar
Anthropic's Claude 3.5 Sonnet surpasses Claude 3 Opus on most benchmarks at half the cost and 2x the speed, while introducing Artifacts for code previews.
Claude 3.5 Sonnet sets new benchmark bar
Anthropic's Claude 3.5 Sonnet surpasses Claude 3 Opus on most benchmarks at half the cost and 2x the speed, while introducing Artifacts for code previews.
Google I/O 2024 — AI everywhere
Google announces Gemini in Search (AI Overviews), Project Astra (real-time multimodal AI), and NotebookLM upgrades at its largest-ever developer conference.
Google I/O 2024 — AI everywhere
Google announces Gemini in Search (AI Overviews), Project Astra (real-time multimodal AI), and NotebookLM upgrades at its largest-ever developer conference.
GPT-4o launched — real-time voice & vision
OpenAI launches GPT-4o ("omni"), enabling seamless real-time voice conversations with emotion detection, singing, and live video analysis. Demoed live on stage.
GPT-4o launched — real-time voice & vision
OpenAI launches GPT-4o ("omni"), enabling seamless real-time voice conversations with emotion detection, singing, and live video analysis. Demoed live on stage.
Llama 3 released by Meta
Meta releases Llama 3 (8B and 70B) as open weights, matching or exceeding many closed models on benchmarks. The 400B model is teased for later release.
Llama 3 released by Meta
Meta releases Llama 3 (8B and 70B) as open weights, matching or exceeding many closed models on benchmarks. The 400B model is teased for later release.
Claude 3 family launched
Anthropic launches Claude 3 with three tiers — Haiku, Sonnet, and Opus — topping benchmarks and offering a 200K context window. Opus beats GPT-4 on MMLU.
Claude 3 family launched
Anthropic launches Claude 3 with three tiers — Haiku, Sonnet, and Opus — topping benchmarks and offering a 200K context window. Opus beats GPT-4 on MMLU.
Gemini 1.5 Pro announced — 1M context window
Google reveals Gemini 1.5 Pro with a 1 million token context window, enabling analysis of entire codebases, long videos, and hour-long audio recordings.
Gemini 1.5 Pro announced — 1M context window
Google reveals Gemini 1.5 Pro with a 1 million token context window, enabling analysis of entire codebases, long videos, and hour-long audio recordings.
Timeline curated from public announcements, press releases, and benchmark reports. Dates reflect public release or announcement dates.