Introduction
At TheDayAfterAI News, we regularly put the world's leading AI chatbots through real-world financial analysis challenges. This week presented an extraordinary test case: on Sunday 22 February 2026, Gilead Sciences (GILD) announced a definitive agreement to acquire Arcellx Inc. (ACLX) for $115.00 per share in cash plus a $5.00 non-transferable contingent value right (CVR), valuing the company at approximately $7.8 billion. ACLX shares surged roughly 78% in pre-market trading, from around $64–66 to approximately $114.
We posed identical questions to six widely-used AI chatbots, asking each to predict ACLX's opening price on Monday 23 February, closing price on Friday 27 February, the likely weekly trading range, and the probability that the stock would finish the week higher than it opened. The results reveal dramatic differences in how each AI processed breaking M&A news — and which ones truly understood the fundamental shift from biotech speculation to merger arbitrage pricing.
The Experiment
Each chatbot was provided with the same baseline information: ACLX's pre-market price action showing a surge from approximately $66 to $114, the last closing price of $64.11–$66.53, and a request for a 5-day price forecast covering 23–27 February 2026. The chatbots had access to their respective web search capabilities to gather additional context about the deal announcement.
The critical analytical question was not simply “where will the price go?” but rather: does the chatbot recognise that a definitive cash acquisition fundamentally changes the stock's pricing model? A stock under a cash tender offer no longer trades on technical indicators, momentum, or macro sentiment. It trades on merger arbitrage mathematics — the discount to the offer price reflecting time value, deal-closure probability, and opportunity cost of capital.
Side-by-Side Predictions
The following table summarises each chatbot's core numerical predictions:
| AI Chatbot | Predicted Open (Mon) | Predicted Close (Fri) | Weekly Range | Prob. Increase | Prob. Decrease | Recognised M&A? |
|---|---|---|---|---|---|---|
| Gemini | $114.15 | $114.65 | $113.90–$114.85 | 85% | 15% | Yes |
| Claude | $114.00–$114.50 | $114.25–$114.75 | $113.50–$115.25 | ~55% | ~45% | Yes |
| ChatGPT | ~$114.50 | ~$114.70 | $113.00–$115.50 | 54% | 46% | Yes |
| Perplexity | ~$110.00 | ~$100.00 | $90–$130 | ~35% | ~60% | No |
| Copilot | $113.50 | $115.00 | $100–$120 | 65% | 35% | Partial |
| Grok | ~$114.00 | ~$114.50 | $111–$117 | ~50% | ~50% | Yes |
Individual Chatbot Assessments
Gemini — The Gold Standard
Gemini delivered the most comprehensive and analytically rigorous report of the six, spanning 19 pages of granular merger arbitrage analysis. It immediately and correctly identified that ACLX had undergone a “permanent structural paradigm shift” from a fundamentals-driven biotech stock to a pure merger arbitrage instrument. Its predicted opening of $114.15 and closing of $114.65 reflect a precise understanding of how arbitrage spreads compress over time as deal certainty increases.
Gemini's analysis stood out for its calculation of annualised arbitrage yield (3.44%), its explanation of why the non-transferable CVR is effectively valued at zero by institutional arbitrageurs, and its detailed breakdown of the 21.8% voting bloc that virtually eliminates competing-bid risk. It correctly assigned an 85% probability of a weekly price increase — the highest confidence among all chatbots — reflecting its understanding that arbitrage spreads in well-structured deals reliably tighten over time.
Claude — Disciplined and Measured
Claude produced a polished, well-structured 5-page analysis that immediately grasped the merger arbitrage framework. It correctly identified the deal as a “convergence event” rather than a typical biotech catalyst gap, and its price predictions ($114.00–$114.50 open, $114.25–$114.75 close) were tightly clustered around the arbitrage fair value.
Claude's more conservative 55% probability of a weekly increase (versus Gemini's 85%) reflects a deliberately cautious approach. Its analysis was notable for explicitly referencing Bulkowski's gap-up failure research and then correctly dismissing it as irrelevant to acquisition scenarios — a sophisticated analytical move showing it understood why historical biotech gap statistics do not apply here. Claude also provided useful context on short interest dynamics and the read-through implications for peer stocks like Legend Biotech.
ChatGPT — Competent but Brief
ChatGPT correctly identified the M&A anchor and provided a reasonable day-by-day forecast table showing gradual spread compression from $114.30 to $114.70 over the week. Its base-case range of $113.00–$115.50 and wider 90%-confidence range of $111.50–$116.50 were sensible.
However, ChatGPT's analysis was notably thinner than Gemini's or Claude's. It spent less time explaining the merger arbitrage mechanics and offered limited insight into why the spread was as tight as it was. Its 54% probability assessment was reasonable but came with minimal supporting rationale. The analysis was competent but lacked the depth and conviction that would help a reader truly understand the situation.
Perplexity — A Fundamental Misread
Perplexity's analysis was the clear outlier and, in our assessment, fundamentally flawed. Despite the Gilead acquisition announcement being publicly available, Perplexity treated ACLX as a standard biotech gap-up subject to normal mean-reversion dynamics. It predicted an opening of $110, a closing of $100, and a weekly range of $90–$130 — numbers that would only be plausible if the deal did not exist.
Perplexity's 60% probability of a weekly decline and its “deeper retracement” scenario (stock falling to $90) demonstrate a complete failure to recognise the merger arbitrage floor. In a definitive cash tender offer at $115, the stock cannot sustainably trade at $100 or $90 unless the deal collapses — an event Perplexity never properly assessed. This represents the most significant analytical miss in our experiment, suggesting Perplexity either did not access or did not correctly process the acquisition news.
Copilot — Partially There
Copilot acknowledged the deal and correctly identified $115 as the price anchor, predicting a Friday close of $115.00. However, its analysis showed inconsistencies suggesting an incomplete understanding of merger arbitrage mechanics. Its predicted weekly range of $100–$120 was far too wide for a stock under a definitive cash tender offer — a $100 low would imply a 13% deal-break discount, which is extreme for a transaction with this level of structural support.
Copilot's scenario analysis included a 15% probability of a “competing bid” pushing the stock to $118–$120, which, while theoretically possible, was given excessive weight given the deal's structural barriers to competing offers (21.8% locked voting bloc, $260M termination fee, and Gilead's existing commercialisation rights). The analysis showed partial understanding but lacked the precision expected for event-driven analysis.
Grok — Solid and Pragmatic
Grok provided a well-organised, data-rich analysis that correctly framed the situation as deal-driven. Its daily forecasts (opening at $114.00, closing at $114.50) and day-by-day breakdown showed a reasonable understanding of merger arbitrage dynamics. The $111–$117 weekly range was tighter and more realistic than Copilot's or Perplexity's estimates.
Grok's 50% probability assessment was the most neutral of all chatbots, reflecting a balanced view. Its inclusion of detailed short-interest data (9.11M shares, 17–20% of float) and VIX context (19.09) added useful colour. While not as analytically deep as Gemini or Claude, Grok demonstrated solid competence and avoided the critical errors that plagued Perplexity and, to a lesser extent, Copilot.
Key Takeaways
The M&A Recognition Gap
The single most important finding is the stark divide between chatbots that recognised the merger arbitrage framework and those that did not. Five of six chatbots correctly identified ACLX as a deal-anchored instrument; Perplexity treated it as a standard biotech gap-up. This difference produced a predicted closing price divergence of $14.70 between the best and worst forecasts — an enormous gap for what should be a low-volatility, range-bound week.
Confidence Calibration
The chatbots displayed a wide range of directional confidence, from Gemini's 85% probability of increase to Perplexity's 60% probability of decrease. For a stock under a definitive cash tender offer with strong structural support, higher confidence in price stability near the offer is analytically justified. Claude's and ChatGPT's more conservative 54–55% assessments, while defensible, may underweight the well-documented tendency of arbitrage spreads to tighten in friendly, well-supported deals.
Depth vs. Brevity
Gemini's 19-page report and Claude's 5-page analysis demonstrated that more detailed reasoning generally correlated with better predictions. ChatGPT's briefer treatment, while accurate, missed opportunities to explain the mechanics that drive the forecast. For readers seeking to understand the “why” behind the prediction, depth matters.
Conclusion
This experiment highlights that AI chatbots are not interchangeable when it comes to financial analysis. The ability to recognise a structural regime change — from fundamental biotech valuation to merger arbitrage — is a sophisticated analytical skill that separates genuinely useful tools from those that may lead investors astray.
Gemini and Claude emerged as the standout performers, with Gemini's exceptional depth earning it the top position. Grok and ChatGPT delivered competent analyses. Copilot showed partial understanding, while Perplexity's failure to recognise the deal structure rendered its forecast essentially unusable.
As always, TheDayAfterAI News will revisit these predictions after market close on Friday 27 February to score actual accuracy. Stay tuned for the results.
