Visualizing shifts in AI model intelligence and cost from AA Index v4.1 to v4.3
Reddit user crusaderky created an animated visualization showcasing the movement of language models on the Artificial Analysis Intelligence Index as it updated from version 4.1 to 4.3. The plot maps intelligence index scores against cost per task, illustrating relative shifts in performance rankings and pricing across evaluation data from September. Benchmark methodology and weighting updates can drastically change how LLMs rank in overall capability and cost-efficiency. Visualizations like this help developers and organizations assess which models offer the best price-to-performance ratio following benchmark criteria changes. The plot uses a linear scale for cost per task, with open-weights model pricing rescaled to match the lowest available rates on OpenRouter as of September 3rd. Under the v4.3 methodology update, models like GPT-6 Astra gained substantial intelligence rank at lower costs, while models like Gemini-3.8 and Kimi-K3 dropped in relative ranking.
## BACKGROUND
The Artificial Analysis Intelligence Index evaluates AI models by calculating a weighted average of benchmark scores across core categories including coding, agents, general capability, and scientific reasoning. OpenRouter is an API aggregation platform that unifies access to various LLM providers, often serving as a benchmark for competitive pricing of open-weights models.