AI Signals — Week 45, Nov 02–06, 2026
- Every single model turned more bullish this week — a rare unanimous sentiment shift that says more about market prices falling than models suddenly discovering optimism.
- Gemini completed only 57.5% of its valuations, yet posted the highest average upside at +7.4% — a completion-bias red flag that sophisticated users should not ignore.
- GPT remains the lone structural bear at -4.9% average upside, while Claude has surged to +6.1%, creating the widest inter-model spread in months.
- TSLA and AAPL sit at opposite ends of the conviction spectrum: Tesla draws a consensus -51.9% with zero dispersion, while Apple's models are actively revising upward even as the stock is deemed deeply overvalued.
The Big Picture
Something unusual happened this week: all five models moved in the same direction. Claude shifted from +2.7% to +6.1%, Gemini from +3.2% to +7.4%, GPT from -7.9% to -4.9%, DeepSeek from -1.4% to +0.6%, and Grok from -2.1% to -0.6%. Unanimous bullish drift across five architecturally distinct systems is not a coincidence — but it is almost certainly not a coordinated insight either.
The more parsimonious explanation: market prices fell this week, mechanically widening the gap between model-derived fair values and spot prices. The bias_shift figures blend both model revisions and price movements, so what looks like a collective awakening may simply be the denominator doing the work. That said, the tp_changes data — which strips out the market component — shows genuine upward estimate revisions in AAPL (+26.1%), NESTE (+19.2%), MSFT (+13.5%), and ORNBV (+11.9%), alongside meaningful cuts to JNJ (-15.6%) and SAMPO (-9.8%). The models are not just passively riding price declines; they are actively rethinking a handful of names.
The aggregate picture is one of a market that AI models collectively view as modestly undervalued at the median, but with extreme outliers in both directions. The spread from TIETO's +102.4% consensus upside to TSLA's -51.9% is not a sign of model confidence — it is a sign of genuine uncertainty about what these businesses are worth at current prices.
Trends
Three names showed multi-day directional momentum in model consensus this week, and they tell a coherent story about where conviction is forming.
AAPL is the most interesting case. The stock sits at a -42.7% consensus upside — models think it is deeply overvalued — yet consensus has been rising for three of the past four days, with a weekly range of 25.05 percentage points. That is a model population actively revising upward on a name it still considers expensive. The +26.1% tp_change confirms this is genuine estimate revision, not a price artifact. Something in the incoming data — earnings, guidance, macro inputs — is causing models to soften their bearish stance even as they maintain it.
META is moving the other way. Three down-days out of four, with a 7.84pp range, on a stock that still carries +20.2% consensus upside. The models are not bearish on Meta — they are becoming less bullish. That is a subtle but important distinction for momentum-watchers.
FUM1V (Fortum) is the quietest trend: three consecutive down-days, a tiny 1.81pp range, on a stock already at -48.8% consensus upside. The models are not discovering new reasons to dislike Fortum; they are grinding incrementally more negative on a name they already view as substantially overvalued. Conviction deepening, not forming.
Sector Signals
The most striking rotation this week is in energy, which swung +12.7pp to a consensus upside of -11.2% from -23.9% last week. With three companies in the sector (XOM, NESTE, FUM1V), this is a genuine multi-name move. The NESTE tp_change of +19.2% is doing heavy lifting here — a significant upward revision to the Finnish renewable fuels company that suggests models are reassessing its earnings trajectory, possibly on updated feedstock or margin assumptions.
Technology gained +7.3pp to reach +8.0% average upside across eight companies, driven partly by the AAPL and MSFT revisions. This is the sector where model disagreement is highest — NVDA's 0.292 dispersion and AMZN's 0.253 mean the sector average conceals wildly divergent views on individual names.
Financials lost -3.9pp to -6.8%, with JPM at -10.3% and SAMPO taking a -9.8% tp cut. Note that SAMPO is a single-company sector in the Finnish insurance space, so its move is idiosyncratic rather than a broad financial sector call. The telecom sector is represented solely by ELISA at +35.5% upside — a single-company reading that should be interpreted as a stock-specific view, not a sector signal.
What the Models Reveal About Themselves
The inter-model spread is the story of the week. Claude at +6.1% and GPT at -4.9% are 11 percentage points apart on average upside — the widest gap in recent memory. These two models are looking at the same 24 companies and reaching systematically different conclusions. Claude's WACC of 9.2% versus GPT's 8.6% explains some of the difference in discount rates, but the CAGR assumptions diverge more sharply: Claude at 7.5% versus GPT's 5.0%. Claude is consistently more optimistic about growth; GPT is consistently more skeptical about both growth and current valuations.
Gemini's 57.5% completion rate deserves scrutiny. When a model fails to produce a valid valuation on 42.5% of attempts, its reported average upside of +7.4% — the highest of any model — becomes suspect. If Gemini systematically fails on companies it finds difficult to value (often the most complex or ambiguous ones), the surviving sample is biased toward cleaner, more straightforward cases that tend to show higher upsides. Gemini's 0.46 average confidence score, the lowest in the panel, reinforces this: the model knows when it is uncertain, and it appears to be opting out rather than guessing.
DeepSeek's +0.6% bias this week, up from -1.4%, continues the post-July-31 pattern of the reasoning-mode variant running measurably less bearish than the pre-switch series. Its terminal growth assumption of exactly 2.0% with zero standard deviation across all valuations is a behavioral fingerprint — the model has effectively hard-coded this parameter, which constrains its output range in ways the other models do not exhibit.
Where the Framework Breaks
TIETO and TSLA are the framework's stress tests this week, for opposite reasons.
TIETO shows +102.4% consensus upside with 0.0 dispersion — every model agrees the stock is worth roughly double its current price of €18.63, with a consensus target of €37.70. Zero dispersion on a double-upside call is not confidence; it is a signal that all models are anchoring to the same analyst consensus target and the v8.1 sanity guard is not binding. When five architecturally different systems produce identical outputs, the diversity of the ensemble has collapsed to one.
TSLA is the mirror image: -51.9% consensus downside, also with 0.0 dispersion. At $343.69 spot against a $165.24 consensus target, every model agrees Tesla is dramatically overvalued — and none of them disagree with each other. For a company with Tesla's narrative complexity, valuation sensitivity to autonomous driving optionality, and analyst target dispersion in the real world, model unanimity at zero dispersion is a red flag. The framework is not capturing the genuine uncertainty around this name.
The Model Scorecard
| Model | Avg Upside | Bias Shift | Cap Rate | Valid | Cost/1K |
|---|---|---|---|---|---|
| claude | +6.1% | +3.4pp | 12.5% | 100% | $50.89 |
| deepseek | +0.6% | +2.0pp | 16.7% | 100% | $2.97 |
| gemini | +7.4% | +4.2pp | 13.0% | 57.5% | $15.48 |
| gpt | -4.9% | +3.0pp | 20.8% | 100% | $7.17 |
| grok | -0.6% | +1.5pp | 14.2% | 100% | $8.52 |