The gap between major model releases keeps shrinking. 2023: once every 73 days 2026 (so far): once every 18 days 🤯

@claudeai @ClaudeDevs @OpenAI @AnthropicAI it's called Pacing the Frontier

@claudeai @ClaudeDevs @OpenAI @AnthropicAI lol

@claudeai @ClaudeDevs @OpenAI @AnthropicAI You forgot to add GPT-6 Sol and luna

@claudeai @ClaudeDevs @OpenAI @AnthropicAI good call out

@claudeai @ClaudeDevs @OpenAI @AnthropicAI It's almost like we're building AI with AI.

However, in terms of pushing the frontier, a model release really only matters if it sets a new SOTA level. The labs could release a new model daily starting tomorrow and it wouldn’t matter if they’re all at the same level of capability. I had Opus 5.5 extrapolate and animate when we would hit a daily new SOTA based on ECI metric, and assuming constant growth (which honestly isn’t really accurate if we hit RSI), we’d hit that in 2032.

@claudeai @ClaudeDevs @OpenAI @AnthropicAI

@claudeai @ClaudeDevs @OpenAI @AnthropicAI Its almost like they are using AI to maybe program better AI?!?!?!?

@claudeai @ClaudeDevs @OpenAI @AnthropicAI Very nice visualization :)

@claudeai @ClaudeDevs @OpenAI @AnthropicAI The competition is really starting to heat up

@claudeai @ClaudeDevs @OpenAI @AnthropicAI and what is CRAZY is that apple likely will stick to once per year for updates, further falling behind.

@claudeai @ClaudeDevs @OpenAI @AnthropicAI At 18 days between releases, I’m still learning the last model’s name when the next one drops 😭

@claudeai @ClaudeDevs @OpenAI @AnthropicAI But now they just finetune it and release it, back then, there were huge differences between the models.

@claudeai @ClaudeDevs @OpenAI @AnthropicAI Where's Gemini 😂

@claudeai @ClaudeDevs @OpenAI @AnthropicAI Just like the Kaiju in Pacific Rim, today we got a double event

@claudeai @ClaudeDevs @OpenAI @AnthropicAI Are they actually getting better. As a coder I don’t think the improvements are that good

@claudeai @ClaudeDevs @OpenAI @AnthropicAI How much of these would have just been patches before

@claudeai @ClaudeDevs @OpenAI @AnthropicAI Soon it'll be every day when your counting Open Weight Models The competition will bring us some really incredible Models IMO

@claudeai @ClaudeDevs @OpenAI @AnthropicAI I used to be able to wait a month, but now I can't even wait a week

@claudeai @ClaudeDevs @OpenAI @AnthropicAI "pace the frontier !" 1 day later : grok 4.7 2 days later : Opus 5.5 3 days later : GPT-6 Sol

@claudeai @ClaudeDevs @OpenAI @AnthropicAI Just wait until starts making improvements on AI... You could have 3 to 4 major revisions in the same day.

@claudeai @ClaudeDevs @OpenAI @AnthropicAI How did you make the Visual animation???

@claudeai @ClaudeDevs @OpenAI @AnthropicAI opus 5.5

And we've already sped past an important realization: "Good Enough LLMs" are here. That's when the economics change. Businesses stop asking, "Which model is smartest?" and start asking, "Which model is good enough for this task—at the right cost, latency, privacy, and deployment footprint?" Once that happens, model selection becomes procurement and routing—not a frontier-model beauty contest. That's going to drive a lot of business-oriented decision-making.

@claudeai @ClaudeDevs @OpenAI @AnthropicAI gonna be every two weeks.

@claudeai @ClaudeDevs @OpenAI @AnthropicAI Funding man & GPUs

The release cadence is accelerating, but METR’s data makes the more interesting point: the capability horizon is accelerating too. METR found AI agents’ 50% task-completion horizon roughly doubled every 7 months historically, and their 2026 update finds the post-2024 doubling time closer to ~89 days. So it isn't just more models, faster. It's potentially more capable agents, faster.

@claudeai @ClaudeDevs @OpenAI @AnthropicAI amazing animation, great work 👏

@claudeai @ClaudeDevs @OpenAI @AnthropicAI Its demand and supply.

@claudeai @ClaudeDevs @OpenAI @AnthropicAI Anthropic is increasingly using more AI for their R&D. This could be why we’re seeing more frequent model releases from them

@claudeai @ClaudeDevs @OpenAI @AnthropicAI That’s a remarkable acceleration, from one major release every 73 days to every 18. Exciting to see how quickly the frontier is moving.

@claudeai @ClaudeDevs @OpenAI @AnthropicAI outpacing the frontier!

@claudeai @ClaudeDevs @OpenAI @AnthropicAI its starting to get annoying, sometimes its the same model rebundled and they make it such a big thing

@claudeai @ClaudeDevs @OpenAI @AnthropicAI Yep, i made a chart to track the gap of models that land on pareto frontier on release day Its shrinking quickly RSI in 3 months

@claudeai @ClaudeDevs @OpenAI @AnthropicAI Looks familiar

@claudeai @ClaudeDevs @OpenAI @AnthropicAI Only going to get faster

@claudeai @ClaudeDevs @OpenAI @AnthropicAI The Chinese are forcing their hand. They used sit longer on finalized models and only released them when they had to.