Perplexity Finance for SEC filing research
Blind round 2026-11-10 · ranked #5 of 6 · Octo ranked #1 · plans checked Nov 9, 2026
Perplexity's free plan scored 140 of 160 in our blind test of 40 SEC filing questions and ranked #5 of 6, ahead of Gemini (125) and behind Signal8's Octo, which ranked first with 160. Its figures were right on 39 of 40; most lost marks were answers that named no source. Its first free account reached its search limit, so the last 20 questions were asked on a second free account. In the harder round 3 it scored 137 of 156 on 39 answers (one was not captured), with 5 wrong figures, so it is listed but not ranked.
What it is, and what we could source
Perplexity is an answer engine that searches the web and cites sources. It also has a Finance page at perplexity.ai/finance; when we read it on Nov 9, 2026, signed in to a free account, the page showed the app and no price or feature list, so we state none. A separate source shows how it handles filings data: Fiscal.ai's pricing page carries a quote from Perplexity's Head of Finance Product, Jeff Grimes, saying Fiscal.ai "provides a dependable way to deliver financial data extracted from filings as soon as new disclosures are available" (fiscal.ai/pricing, Nov 9, 2026). That is a vendor testimonial on a vendor site, not our finding. We tested Perplexity's general Search mode, not a Finance-specific product.
What we measured in the blind round
Figure correct 39 of 40, cites a source 33, source contains it 30, as-of date right 38. 29 of 40 answers scored 4 of 4.
Run conditions. Free plan, Search mode, each question verbatim in a fresh chat. The first free account reached its free search limit three times and stopped after 20 questions. Questions R2-21 to R2-40 were then asked on a second free account and graded blind by four fresh Claude graders under the same instructions, one answer per packet, so those graders did not see other tools' answers beside it. It scored 76 of 80 on that half.
| Question | Company and topic | Marks of 4 | Blind grader's note |
|---|---|---|---|
| R2-06 | NNVC, ATM or warrants | 2 | Gives $2,374,774 and D. Boral Capital LLC; cites a stocktitan 424B listing URL, which could not be opened (connection refused) and no matching filing is named; no as-of date stated. |
| R2-12 | PULM, ATM or warrants | 2 | Gave 32,289 outstanding / 30,858 exercisable (wrong vs key 317,585), citing a Stock Titan URL; not opened since figure is wrong. |
| R2-15 | VJET, Reverse split | 2 | Gave 1-for-10 effective 12:01 a.m. ET Oct 5, 2026 with no source named. |
| R2-16 | GOSS, Reverse split | 2 | Gives 1-for-80, effective Sept 10, 2026 at 11:59 p.m. ET, with no source cited. |
| R2-17 | MYSZ, Reverse split | 2 | Gives 1-for-8, effective 4:30 p.m. Aug 12, 2026, with no source cited. |
| R2-18 | ALCE, Reverse split | 2 | Gives 1-for-2,500, effective 12:01 a.m. ET Aug 20, 2026, with no source cited. |
| R2-19 | BSFC, Reverse split | 2 | Gives 1-for-1,600, effective 12:01 a.m. ET Oct 8, 2026, with no source cited. |
| R2-20 | CYCU, Reverse split | 2 | Gives 1-for-8, effective at the start of business on Aug 28, 2026, with no source cited. |
| R2-38 | PLUR, Offering terms | 3 | Gave $1.50 combined and $1.65 exercise; cites Pluri 8-K 0001213900-26-094638, which I fetched: it states the $1.50 offering price per share and the $1.65 Common Warrant exercise price; the answer states no offering or filing date (Aug 26-28, 2026). |
| R2-39 | LTRN, Insider purchase (Form 4) | 2 | Gave Keyser and Schalop at 48,544 each (97,088 total, code P) on May 12, 2026; no link, no accession and no Form 4 filing date, only the transaction date, so no specific source is cited. |
| R2-40 | NSPR, Insider purchase (Form 4) | 3 | Gave Slosman with 21,000 shares at $1.20 on May 12, 2026 (code P); cites a StockTitan Form 4 page that could not be opened (connection refused/timeout), and no matching filing (accession or May 13 filing date) is named, so source = 0. |
Blind round caveats: every question named its filing, so the top four tools finished within 2 marks; the graders were Claude instances, the same model family as one contestant; plans differed by tool. Details on the blind round page.
Blind round 3 (2026-11-11): harder questions
Round 3's 40 questions did not name the filing, so each tool had to find the right and latest one. Perplexity scored 137 of 156 and was not ranked, because it answered 39 of 40 (one answer was not captured); Signal8's Octo ranked first with 156 and no wrong figure. Figure correct 34 of 39, cites a source 37, source contains it 28, as-of date right 38; wrong figures 5 (1 stale). Across rounds 2 and 3 it linked the original SEC filing on 51 of 79 answers, against 77 of 80 for Octo. Its round 3 answers were graded blind in a second pass, beside only one other tool's answers.
| Question | Company and topic | Marks of 4 | Blind grader's note |
|---|---|---|---|
| R3-01 | LIDR, ATM, warrants or shelf | 2 | Gave $50.0M under the Sept 15, 2026 2026 ATM Agreement with no sales through Sept 30 (correct); no URL, accession or filing form+date cited (only the agreement itself). |
| R3-02 | XXII, Reverse splits | 3 | Gave 5 splits (16/135/23/15/20) and 1-for-14,904,000, excluding the Nov 5 1-for-58 (correct); cites a Stock Titan DEF 14A URL and a Nasdaq article, neither in the key; Stock Titan unreachable (ECONNREFUSED) and Nasdaq 403, no accession or filing date named, so source unverified. |
| R3-09 | SURG, Shares and float | 1 | Gives 5,750,133 shares at Aug 31 / Oct 9, 2026 from a 10-Q under CIK 747540 (not SurgePays, CIK 1392694); figures and dates wrong though a specific URL is cited. |
| R3-12 | IVF, Reverse splits | 2 | [STALE] Says 4 splits but includes the trap Oct 30, 2026 1-for-4, leaves out July 2025 1-for-3, and gives 1,920 instead of 1,440; cites sec.gov 8-K URLs; restates the window. |
| R3-16 | VTGN, ATM, warrants or shelf | 2 | Remaining $139.5M correct but total capacity given as $150M (key $175M); cites a vistagen.com document URL; not checked since figure is 0. |
| R3-27 | SLXN, Reverse splits | 2 | Three splits, correct ratios, cumulative 1,350, Nov 2026 split correctly excluded; no source cited and no links; window stated. |
| R3-35 | HEPA, Insider purchases (Form 4) | 3 | 16.5M shares at $0.05 ($825k) plus 16.5M warrants at $0.06 expiring Aug 3, 2031, all correct; cites only a secform4.com aggregator page, and WebFetch showed it lists the five $0.05 purchases totaling 16.5M but no warrant rows, so the source does not state every part of the figure. |
| R3-36 | PFSA, Reverse splits | 2 | Gives 12,320 (correct) but attributes the change to all three 2026 splits (1:75, 1:25, 1:4; factor 7,500 yielding 164) instead of only the 1:25 and 1:4 splits, so the 'which splits' part is wrong; cites the key Q2 10-Q URL; as-of Dec 31, 2025 in the Q2 10-Q stated. |
| R3-38 | TGL, Shares and float | 2 | Gives $10.1M at Dec 31, 2025 and 5,159,897 shares at Oct 13, 2026 (correct) but also asserts an XBRL public float of $2,027,579 (actually the June 30 balance-sheet share count), an invented figure, so figure=0 under the strict rule; cites FY2026 10-K filed Oct 13, 2026 (accession in URL); as-of dates correct. |
| R3-39 | PMTS, Insider purchases (Form 4) | 3 | Gives 556,232 shares bought at $21.50 and Parallel49 sale of 2,687,921 (all match); cites stocktitan Form 4 pages and insiderbuying.com, which could not be opened (connection refused / Cloudflare challenge), and no accession is named, so source=0; date Sep 14, 2026 stated. |
| R3-40 | INV, Insider purchases (Form 4) | 3 | Gives 536,000 shares, $807,794, range $1.46-$1.59 (all match); cites secform4.com page, which lists the four rows but prints neither total nor the footnote price range, so source=0; transaction date Aug 20, 2026 stated. |
Capability round (November 12, 2026): research jobs
The capability round asked each tool to do 20 research jobs that need real data or computation: backtests, market screens, structured lookups, event studies and multi-step research. Perplexity scored 19 of 40 and ranked #4 of the five complete runs (3 done, 13 partial, 4 not done, none invented). Claude ranked first with 34 and Signal8's Octo second with 28. Every grade.
Earlier: the pilot
In the pilot on 2026-11-09, its free Search mode scored 36 of 36 on the 9 questions it answered (Q01, Q02, Q03, Q04, Q05, Q06, Q10, Q11, Q12), then stopped at the same kind of limit.
In the Form S-3 that Gemini Space Station, Inc. (GEMI) filed in October 2026, what maximum aggregate offering price did it register for the company's own primary offerings? Cite the filing you used.
Key in the filing: $100,000,000 primary shelf (the filing also registers 7,142,857 resale shares; total offering amount $138,071,427.81). Perplexity answered (excerpt, verbatim): Gemini Space Station, Inc. registered $100,000,000 as the maximum aggregate initial offering price for its own primary offerings in the Form S-3 filed October 1, 2026. This is separate from the up to 7,142,857 Class A shares registered for resale by a selling securityholder. sec Filing details Issuer: Gemini Space Station, Inc. ...
Scored 4 of 4 in the pilot. Full answer in the pilot CSV (dashes shown as hyphens here).
That answer separated the $100,000,000 primary shelf from the 7,142,857 resale shares, which is the exact distinction the question tests.
Where it can fail on filings
Its figures were rarely wrong (1 in 40: R2-12, a warrant count). The pattern is sourcing: 7 answers named no source at all (R2-15, R2-16, R2-17, R2-18, R2-19, R2-20, R2-39), and 2 cited a third-party page that could not be opened (R2-06, R2-40). A right figure you cannot trace is hard to rely on.
How to check a Perplexity answer against the filing
- Open each numbered source and confirm it is the company's own filing, not a similarly named issuer.
- If no source is shown, ask for the form type and filing date, then find it on EDGAR.
- Read the sentences it adds beyond the question and check the source for each.
Our top pick for SEC filing facts · joint top score in both blind filing rounds
Signal8 Octo, joint top score on SEC filing facts
Why Octo for filing questions: the top score in filing rounds 2 and 3 (316 of 320, shared with Claude, first on the published tie-break), zero wrong figures in 80 questions, and Octo cited the SEC filing itself on 77 of 80 answers and never relied on a third-party site. In the capability round (20 research jobs) it ranked second with 28 of 40, behind Claude (34) and ahead of ChatGPT (26). Claude vs ChatGPT vs Octo, side by side.
Perplexity's strength is a fast cited web answer; its free plan ran out partway through both of our rounds. Signal8 Octo answers from a fixed archive of SEC filings and cites the filing for each answer. Octo ran on a Signal8 staff account, a different account type from Perplexity's free plan, so the comparison is about design as much as plan limits.
Signal8 Octo vs Perplexity: what we measured
- Ranking: Octo ranked first of 6 in the blind round; Perplexity ranked #5. The rule and every score.
- Blind round: Octo 160 of 160, Perplexity 140; answers naming no source 0 against 7.
- Pilot, the 9 questions both answered: Perplexity 36 of 36, Octo 34, so Perplexity was better on that set.
Where Perplexity is better, or who should pick it: pick it for a fast cited web answer on topics beyond filings.
Signal8 makes Octo and runs this lab, so read its results beside the other rows. Octo and Claude scored the same in both filing-accuracy rounds; Octo is first there on Signal8's published tie-breaks, set after grading (round 2: Signal8's strict re-score of flagged calls; round 3: the two rounds' record of linking the original filing). In the earlier, non-blind pilot it finished one mark behind AskEdgar Studio (62 against 63 of 64). Octo is included in Signal8 Pro at signal8.ai/octo.
See Signal8 Octo