Mar 2022🎨Midjourney v1
Midjourney launches its first version in open beta via Discord. Introduces text-to-image generation to mainstream users.
A chronological reference of every major AI model release and milestone — from ChatGPT's launch to GPT-5's announcement. 50+ releases across text, image, video, and audio AI.
Midjourney launches its first version in open beta via Discord. Introduces text-to-image generation to mainstream users.
OpenAI releases DALL-E 2, a major leap in image quality, capable of realistic edits and outpainting. Access via waitlist.
Stability AI releases Stable Diffusion v1.4 with open weights. First major open-source image model — anyone can run it locally for free.
OpenAI launches ChatGPT on November 30, 2022. Reaches 1M users in 5 days, 100M in 2 months — fastest-growing app ever. Powered by GPT-3.5 Turbo.
OpenAI releases GPT-4 — the first multimodal GPT (vision + text). Dramatically improved reasoning, passes bar exam at 90th percentile.
Anthropic launches Claude 1 via API. Introduces Constitutional AI and focuses on safety and helpfulness over raw capability.
Midjourney v5 sets a new standard for photorealistic AI images. Hands and fine details dramatically improved.
Meta releases Llama 2 with open weights — 7B, 13B, and 70B models. First commercial-grade open-source LLM family.
Stability AI releases SDXL — significantly improved image quality with 1024×1024 resolution and dual-encoder architecture.
Mistral AI releases Mistral 7B — outperforms Llama 2 13B on benchmarks with only 7B parameters. Open weights, Apache 2.0 license.
Google releases Gemini — a family of multimodal models (Ultra, Pro, Nano). Gemini Pro launches in Bard. Positions as direct GPT-4 competitor.
Midjourney v6 launches with dramatically improved photorealism, better text rendering in images, and more accurate prompt following.
OpenAI announces Sora — a text-to-video model capable of generating 1-minute photorealistic videos. Shown only to researchers initially.
Anthropic releases Claude 3 Opus, Sonnet, and Haiku simultaneously. Opus outperforms GPT-4 on multiple benchmarks at launch. All support 200K context.
Google releases Gemini 1.5 Pro with 1M (later 2M) token context window — enough for entire codebases and long-form video analysis.
Meta releases Llama 3 (8B and 70B). The 70B outperforms Llama 2 70B by large margins. Fine-tunes become state-of-the-art open models.
OpenAI releases GPT-4o ("omni") — real-time audio/vision/text model. Demo shows natural voice conversations with emotional awareness. Free tier gets GPT-4 level.
Anthropic releases Claude 3.5 Sonnet — outperforms Claude 3 Opus at a fraction of cost. Becomes the go-to coding model. Introduces Artifacts feature.
Kuaishou launches Kling AI — a video generation model capable of up to 2-minute videos with strong physics simulation. Competitive with Sora.
OpenAI releases GPT-4o mini — replaces GPT-3.5 Turbo at 60% lower cost. Cheaper than Claude Haiku. Strong multimodal support.
Black Forest Labs (former Stable Diffusion team) releases Flux 1 — three variants: Pro, Dev (open), and Schnell. Surpasses SDXL in quality.
OpenAI releases o1 ("Strawberry") — a reasoning model that thinks step-by-step before answering. Dramatically better at math, coding, and science benchmarks.
Anthropic officially releases Claude 3.5 Haiku as their fastest and most affordable model, with 200K context. Outperforms Claude 3 Opus on some benchmarks.
Google releases Gemini 2.0 Flash — successor to 1.5 Flash with better performance and a generous free tier. Supports 1M context and multimodal I/O.
OpenAI releases o3-mini — an efficient version of the o3 reasoning architecture, available at lower cost than o1. Configurable reasoning effort levels.
OpenAI releases GPT-4.1 — improved instruction following, longer 1M context window, and better coding performance. API-first release.
OpenAI releases o3 — a full reasoning model with frontier-level performance on AIME math, SWE-bench coding, and ARC-AGI. Available to Pro users.
Google releases Gemini 2.5 Pro — Google's most capable model with strong reasoning, 1M context, and top scores on MMLU Pro and MATH benchmarks.
Anthropic releases Claude 4 Sonnet — the next generation of Anthropic's flagship model. Improved agentic capabilities, coding, and extended thinking.
Google's most powerful Gemini model — positioned to compete with GPT-5. Expected to have 2M+ context window and advanced multimodal reasoning.
OpenAI has signaled the development of GPT-5 — expected to combine reasoning (o-series) and GPT-4o's multimodal capabilities into one frontier model.
ChatGPT was publicly released by OpenAI on November 30, 2022. It reached 1 million users in 5 days and 100 million users in just 2 months.
GPT-4 was released on March 14, 2023. It introduced image understanding and dramatically improved reasoning compared to GPT-3.5.
Stable Diffusion v1.4 was publicly released on August 22, 2022, with open weights — allowing free local installation.
Key 2024 releases: Claude 3 family (Mar), Sora announcement (Feb), GPT-4o (May), Llama 3 (Apr), Claude 3.5 Sonnet (Jun), GPT-4o mini (Jul), Flux 1 (Aug), and o1 reasoning model (Sep).