GPT-4
*Multimodal generalist that reset 2023 expectations for instruction-following and tool use at scale.*
Shipped by OpenAI · · class: frontier-llm
At its March 2023 introduction, OpenAI positioned GPT-4 as a step-change beyond GPT-3.5 on reasoning, exam-style tasks, and multimodal inputs—with comparatively little public detail on architecture and training compute. Third-party replication of “GPT-4 class” behavior became an organizing goal for the industry.
Since release, the honest story is gradual capability creep via post-training, tooling (function calling, later agents), and eval evolution—not a single static capability number. Benchmark leadership rotated as competitor families shipped; the durable observation is distribution: GPT-4-era behavior became the default bar for enterprise pilots.
Tracked claims about this model
- Everybody got super excited last year, but we are running out of improvements; at least for a little… [refuted]
- GPT-5 could be a significant leap forward, but there's still a lot of work to do on it. [weakening]
- The most important thing and it sounds like the most boring thing I can say... the models are just g… [refuted]
- OpenAI will be cash-flow positive by 2024. [refuted]
- AI could boost labor productivity growth across the developed world by about 1.5 percentage points p… [open]