Somewhere around the ninth “OpenAI just announced” Slack message in a single week, I gave up trying to hold the timeline in my head. Model names stopped mapping to release dates, release dates stopped mapping to what actually changed, and “the new one” became ambiguous across at least four labs simultaneously. That’s not a personal failing so much as a description of the last two years: the industry went from a handful of frontier labs shipping a model every few months to dozens of labs (frontier and open-weight alike) shipping something worth knowing about most weeks.

So I built the timeline I wished existed: every model below is placed in the era it actually shaped, carries the specs a developer would check before reaching for it (parameters, context window, license, how to run it), and links back to a first-party source. Nothing here is guessed. Where a spec wasn’t publicly disclosed, the card just doesn’t show it, rather than making one up.

Four eras structure the scroll:

• The Chat Era (through August 2024): scale and instruction-tuning as the whole story, one-shot assistants competing on breadth.
• The Reasoning Turn (September 2024 to June 2025): test-time compute becomes a first-class feature, chain-of-thought moves from prompt trick to trained behavior.
• The Agentic Era (July 2025 to February 2026): long-horizon tool use, coding agents, and multi-step planning become the headline capability.
• Efficiency & Hybrids (March 2026 onward): sparse MoE and linear-attention hybrids push frontier-adjacent capability down into models small enough to run locally.

Browse chronologically within an era, or jump straight to one from the rail below the search bar. Click any card for the full writeup: specs, how to access it, and the one sentence on why it mattered. The search box on top isn’t just a text filter either: it understands org:, era:, license:, open:true, and numeric filters like params:>100b, so you can slice the whole list by whatever axis you actually care about.