The Big Picture

This was a week of remarkable, if uncomfortable, agreement. Point every model at the same 23 companies and ask what they're worth — and all five come back bearish. Not mildly cautious. Bearish. GPT logged an average upside of -15.3%, DeepSeek sat at -10.3%, Grok at -8.4%, Gemini at -6.1%, and even the most optimistic reader in the room, Claude, still found -2.9% of average downside across its valid valuations.

The consensus target prices tell a stark story. AAPL at $263 spot, with a model consensus target of $176. XOM at $150 with a target of $87. JPM trading at $299 against a DCF-implied $207. These aren't minor valuation quibbles — these are structural calls that the market has priced in growth that the models' cash flow arithmetic simply cannot support at current discount rates averaging 9.1–9.7% WACC across the panel.

The constructive read: AI models use DCF frameworks that are inherently skeptical of momentum, and US large-cap equities have been momentum-driven for years. The troubling read: when five independent architectures all reach the same directional conclusion, it at minimum demands that investors understand why — rather than dismissing it as a quirk of quantitative methodology.

What the Models Reveal About Themselves

The most instructive personality differences this week are not in the outputs but in the structural DNA of how each model reasons.

Gemini and Grok are the optimists of the cohort — highest CAGR estimates at 8.3% and 8.5% respectively — yet they carry the highest cap rates (42.6% and 43.5%), meaning the platform's guardrails are catching a meaningful share of their terminal value assumptions as potentially excessive. They dream big, but the framework keeps a hand on their shoulder.

DeepSeek is the week's sleeper story. Perfect 100% validity, the tightest terminal growth dispersion (0.28 standard deviation), and a confidence score of 0.65 — higher than both Claude and GPT. It is also, extraordinarily, 16x cheaper than Claude per thousand valuations at $2.03 vs $32.08. Whether that price gap reflects genuine capability differences or simply API pricing strategy is a question worth sitting with.

Claude, meanwhile, shows the widest uncapped terminal value deviation at 73.9% — suggesting it's generating the most aggressive raw growth assumptions before the cap mechanism intervenes. High aspiration, heavily disciplined.

Where the Framework Breaks

GPT's terminal growth rate deserves its own paragraph, because it is not a valuation characteristic — it is a confession. Across 63 valid valuations spanning technology, energy, financials, pharmaceuticals, and industrial companies spread across two continents, GPT applied a terminal growth rate of exactly 2.00% in every single case. Standard deviation: 0.0. Precisely zero.

Terminal growth rates should vary. A Finnish telecom utility like ELISA and a hypergrowth AI infrastructure company like NVDA should not share an identical long-run nominal growth assumption. The fact that they do here means GPT is not estimating terminal growth — it is substituting a hard-coded default and presenting it as analysis. This matters enormously for DCF outputs, because terminal value typically represents 60–80% of the total valuation. A model that treats this parameter as a constant is, in a meaningful sense, not doing a DCF at all. It is doing an EBIT-to-perpetuity calculation wearing a DCF costume.

GPT and Grok also both triggered terminal value caps on 3 valuations each — the only models to do so — suggesting that when they do vary from their defaults, they occasionally overcorrect dramatically in the opposite direction.

The Model Scorecard