Via pngplay.com
BofA's new tool benchmarks leading large language models by performance and per-token pricing, with Anthropic's Claude family scoring highest on intelligence.
Wall Street is watching the AI arms race more closely than ever, and Bank of America just built a scoreboard. The bank has launched an internal AI model tracker that evaluates leading large language models across two dimensions that actually matter to enterprise buyers: how smart the model is, and how much it costs to run.
The tracker places Anthropic’s Claude family at the top of its intelligence index, with scores around 63 for the Opus and Sonnet variants. Those models also rank high on usage intensity, meaning organizations running Claude are spending more, likely because the performance justifies it.
What the tracker actually measures #
The tool benchmarks LLMs on a composite intelligence score while simultaneously tracking per-million-token pricing and GPU rental rates, the two cost levers that determine whether an AI deployment is financially viable at scale.
Per-token pricing is essentially the meter running every time a model reads or writes text. At high volumes, even small differences in price per million tokens compound into meaningful budget line items. GPU rental rates, meanwhile, determine what it costs to fine-tune or host models privately, a consideration for banks handling sensitive client data.
The tracker shows both metrics are in flux as of mid-2026, reflecting an AI infrastructure market that hasn’t settled into stable pricing.
No formal public announcement has accompanied the tracker’s launch, and Bank of America has not published its methodology. The analysis appears to have originated from a client-facing research report rather than a press release, which explains why specific technical details remain inside the walls of BofA’s institutional client base.
BofA is not just watching AI, it’s running it #
Bank of America currently operates more than 270 AI and machine learning models internally, with over 300 approved AI use cases across its business lines. It processes roughly 400,000 AI prompts per day, a workload that would make the bank itself one of the larger enterprise AI consumers by volume.
At that scale, a one-dollar-per-million-token price difference translates into real money. The tracker isn’t just a research product for clients; it’s also a procurement intelligence tool for BofA’s own AI operations.
What this means for the AI market #
Anthropic’s position at the top of the intelligence ranking carries weight coming from a financial institution with BofA’s analytical resources. A systematized benchmark from a major bank introduces a more structured form of due diligence into a market that has mostly relied on benchmark cherry-picking from the model developers themselves.
For the broader AI infrastructure market, the tracker’s focus on GPU rental rates adds a layer of transparency to a segment that has operated with relatively opaque pricing. BofA putting numbers to that market, even in a client-only format, starts to establish reference points that didn’t exist before. Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our