kwj.ai · acquisition inquiries from >$999view prospectus →
The Domesday Book ofKWJ · AI

Era IV · 2022–2023

The Consumer-LLM Bloom

By C.W. Jameson · Published 19 May 2026 · Last reviewed 19 May 2026

In the year following ChatGPT's release the field went from one major commercial model to four. GPT-4, Claude 1 and 2, Bard, and Llama all arrived inside twelve months. Pricing dropped by an order of magnitude. The first generation of LLM-native startups raised at meaningful valuations. Operators got used to switching providers monthly.

GPT-4

Released 14 March 2023. The technical report was famously thin on architectural detail. The capability was substantial — bar exam, AMC, AP exams, all comfortably above the human percentile. The cost was punitive at launch and dropped repeatedly through the year.

Claude 2

July 2023, with the then-unprecedented 100K token context window. Operators who had been splitting prompts across calls suddenly had room. The voice of the model (calmer, hedged, more deliberate) became a recognisable trait.

Llama 2

July 2023, Meta's first commercially-licensed open-weights release. The Llama 2 chat fine-tunes set the open-weights baseline for the year that followed. Every serious open-weights derivative until Llama 3 traced back to it.

Signature models of the era

  • GPT-4
  • Claude 2
  • Bard / Gemini 1
  • Llama 2

Technical shifts

  • Context windows pass 100K tokens
  • Open-weights catch within striking distance of frontier on instruction-following
  • Function-calling APIs ship across all major providers

Market shifts

  • The first $100M+ revenue LLM-native startups
  • Anthropic raises at $4.1B; OpenAI raises at $29B
  • The 'race to zero' on input pricing begins

Authentication — is the document from this era?

TellMeaning
Mentions GPT-4 by name without a version suffixDocument dates to 2023 or early 2024 when 4 was unmodified.

Agents catalogued in this era

  • Mistral 7BThe model that proved 7B parameters could beat 13B models. Reshaped expectations for small open models.
  • Claude 2Anthropic's first publicly competitive release. 100K context when GPT-4 had 8K.

Primary sources

  1. [1] OpenAI: GPT-4 Technical Report2023-03-15
  2. [2] Anthropic: Claude 22023-07-11

From the Almanac shop

The AI Eras — Pocket Field Guide

Ten eras of AI on a single foldable. The Almanac in your pocket.

$19Coming soon

Back to the timeline