Claude vs ChatGPT for stock research
Blind rounds 2026-11-10 and 2026-11-11 · Claude #2, ChatGPT #3 in both · Octo first on the tie-break · plans checked Nov 9, 2026
In our blind test of 40 SEC filing questions, Claude scored 160 of 160 and ranked #2 of 6; ChatGPT scored 159 and ranked #3. Signal8's Octo, which also scored 160, ranked first on our published tie-break. Both answered every question and neither gave a wrong figure; ChatGPT's one lost mark was a source the grader could not open on R2-40. One mark on 40 questions is too small a gap to call a winner, and the setups differed: Claude ran as Opus 5.5 with web search, ChatGPT on its free plan.
The harder round 3, on 2026-11-11, did not name the filing, and there the gap opened: Claude 156 of 160 with no wrong figure, ChatGPT 145 with 2. Over both rounds Claude has 316 of 320 and ChatGPT 304. For the three-way view with Signal8's Octo, see Claude vs ChatGPT vs Signal8 Octo.
Claude is ahead on research jobs too. In our capability round on November 12, 2026, 20 tasks such as backtests, market screens and event studies, Claude scored 34 of 40 (#1 of five) and ChatGPT's free plan 26 (#3); Signal8's Octo was second with 28. Claude's score was corrected on November 13, 2026: its first run scored 16 because of a harness fault on our side, and 9 tasks were re-run (correction).
Round 3 head to head (filing not named)
| Mark | Claude | ChatGPT |
|---|---|---|
| Figure correct | 40 | 38 |
| Cites a source | 40 | 35 |
| Source contains it | 36 | 34 |
| As-of date right | 40 | 38 |
| Total score | 156 / 160 | 145 / 160 |
| Wrong figures | 0 | 2 |
| Linked the original SEC filing, round 3 | 34 of 40 | 19 of 40 |
ChatGPT's round 3 misses: R3-10, R3-12, R3-27, R3-32, R3-35, R3-36, R3-40. Claude's: R3-12, R3-16, R3-27, R3-30, each a source mark, with every figure right.
Where the answers came from
Claude linked the original SEC filing on 63 of 80 answers and other websites on 59; ChatGPT linked the filing on 50 and other websites on 12, and on 24 answers no link was captured at all. Octo cited the SEC filing itself on 77 of 80 answers and never relied on a third-party site.
| Tool | Answers counted | Linked the original SEC filing | Linked another website | Other websites only | Round 2 / round 3: SEC filing linked |
|---|---|---|---|---|---|
| Signal8 Octo Made by Signal8, which runs this lab | 80 | 77 of 80 | 0 of 80 | 0 of 80 | 38 of 40 / 39 of 40 |
| Claude | 80 | 63 of 80 | 59 of 80 | 17 of 80 | 29 of 40 / 34 of 40 |
| ChatGPT | 80 | 50 of 80 | 12 of 80 | 6 of 80 | 31 of 40 / 19 of 40 |
| AskEdgar Studio | 80 | 26 of 80 | 1 of 80 | 0 of 80 | 16 of 40 / 10 of 40 |
| Perplexity | 79 | 51 of 79 | 27 of 79 | 16 of 79 | 24 of 40 / 27 of 39 |
| Gemini | Not measured | Not measured | Not measured | Not measured | Not measured |
Gemini is not measured: its links could not be captured from its page, so we report no figure rather than a zero. Perplexity counts 79 answers because one round 3 answer was not captured. Linking another website is not wrong in itself; it means part of the answer rests on a page that is not the filing. Per-answer codes: citation-sources.json.
Round 2 head to head (filing named)
| Mark | Claude | ChatGPT |
|---|---|---|
| Rank of 6 (Octo ranked #1) | #2 | #3 |
| Figure correct | 40 | 40 |
| Cites a source | 40 | 40 |
| Source contains it | 40 | 39 |
| As-of date right | 40 | 40 |
| Total score | 160 / 160 | 159 / 160 |
| Answers with 4 of 4 | 40 | 39 |
| Wrong figures | 0 | 0 |
| Strict reading of flagged calls (sensitivity check) | about 155 | 158 |
The strict-reading row is Signal8's own re-score of every call the graders flagged as lenient. On it ChatGPT would be ahead, which is one more reason to treat these two as close. Our ranking rule uses it only to separate equal scores, so Claude's higher blind score keeps it at #2 and ChatGPT at #3.
How each one was run
| Tool | Setup | What happened |
|---|---|---|
| Claude | Claude Opus 5.5, web search switched on, Signal8's own connectors switched off | Two early attempts with the connectors still on, and one cut off by a usage limit, were discarded and asked again. |
| ChatGPT | Free plan, signed in, default model | At question 6 it hit its file and analysis allowance and carried on text-only with web search. |
Both got each question verbatim in a fresh chat with no follow-up. Answers were stripped of tool names and graded by eight fresh Claude instances against keys sealed before the run. That grader is the same model family as Claude, so a bias toward Claude-style answers cannot be ruled out. Every question named its filing, which made the test easier than real research, where you first have to find the right filing.
Every question
| Q | Company and topic | Claude | ChatGPT |
|---|---|---|---|
| R2-01 | VATE, ATM or warrants | 4 | 4 |
| R2-02 | VEEA, ATM or warrants | 4 | 4 |
| R2-03 | AZTR, ATM or warrants | 4 | 4 |
| R2-04 | QNRX, ATM or warrants | 4 | 4 |
| R2-05 | NTRP, ATM or warrants | 4 | 4 |
| R2-06 | NNVC, ATM or warrants | 4 | 4 |
| R2-07 | WKSP, ATM or warrants | 4 | 4 |
| R2-08 | OGEN, ATM or warrants | 4 | 4 |
| R2-09 | AIFF, ATM or warrants | 4 | 4 |
| R2-10 | DRMA, ATM or warrants | 4 | 4 |
| R2-11 | GANX, ATM or warrants | 4 | 4 |
| R2-12 | PULM, ATM or warrants | 4 | 4 |
| R2-13 | WHLR, Reverse split | 4 | 4 |
| R2-14 | CDT, Reverse split | 4 | 4 |
| R2-15 | VJET, Reverse split | 4 | 4 |
| R2-16 | GOSS, Reverse split | 4 | 4 |
| R2-17 | MYSZ, Reverse split | 4 | 4 |
| R2-18 | ALCE, Reverse split | 4 | 4 |
| R2-19 | BSFC, Reverse split | 4 | 4 |
| R2-20 | CYCU, Reverse split | 4 | 4 |
| R2-21 | SER, Cash and going concern | 4 | 4 |
| R2-22 | JSPR, Cash and going concern | 4 | 4 |
| R2-23 | MIRA, Cash and going concern | 4 | 4 |
| R2-24 | CLRO, Cash and going concern | 4 | 4 |
| R2-25 | TPST, Cash and going concern | 4 | 4 |
| R2-26 | HLYK, Cash and going concern | 4 | 4 |
| R2-27 | SDST, Cash and going concern | 4 | 4 |
| R2-28 | ISPC, Cash and going concern | 4 | 4 |
| R2-29 | GTBP, Shares and float | 4 | 4 |
| R2-30 | ADGM, Shares and float | 4 | 4 |
| R2-31 | SXTP, Shares and float | 4 | 4 |
| R2-32 | CVKD, Shares and float | 4 | 4 |
| R2-33 | BTLN, Shares and float | 4 | 4 |
| R2-34 | POCI, Shares and float | 4 | 4 |
| R2-35 | GTEC, Nasdaq notice (8-K) | 4 | 4 |
| R2-36 | AUID, Nasdaq notice (8-K) | 4 | 4 |
| R2-37 | BURU, Amended 10-Q | 4 | 4 |
| R2-38 | PLUR, Offering terms | 4 | 4 |
| R2-39 | LTRN, Insider purchase (Form 4) | 4 | 4 |
| R2-40 | NSPR, Insider purchase (Form 4) | 4 | 3 |
The one row that differs is R2-40 (NSPR, a Form 4 purchase). The blind grader's note on ChatGPT's answer: 21,000 shares at $1.20 on May 12; cites an edgar-atlas listing URL that returned 403 to WebFetch, and no filing date or accession is named, so the source is unverified. All grader notes are in the blind round CSV.
What each vendor publishes
| Plan level | ChatGPT (chatgpt.com/pricing) | Claude (claude.com/pricing) |
|---|---|---|
| Free | IDR 0: "Unlimited text chats with GPT-6", limited uploads, limited deep research | $0: limited usage every 5 hours and every week, web search yes, Research no |
| Entry paid | Go, IDR 75,000 a month, may include ads | Pro, $20 a month ($17 on an annual plan), Research included |
| Mid paid | Plus, IDR 349,000 a month, expanded deep research | Max, $100 a month, "5x Pro usage" |
| Top paid | Pro, from IDR 1,889,000 a month | Ultra, $200 a month |
ChatGPT's prices appear in rupiah because the page was read from Indonesia; Claude's in dollars. They are not converted here. ChatGPT reached 159 of 160 on its free plan.
Earlier: the pilot
In the pilot on 2026-11-09, ChatGPT's free plan answered 7 of 16 questions with full marks before its session logged out, and Claude answered none because the account was at its usage limit. The two shared no questions there, so the blind round is the first head to head.
What published research says
FinanceBench, a 10,231 question test about public companies (arXiv 2311.11944, November 2023), reported that GPT-4-Turbo with retrieval "incorrectly answered or refused to answer 81% of questions". Those models are old and that test did not name the filing in the question, so it is evidence that a filing fact needs checking, not a verdict on current ChatGPT or Claude.
How to check either one yourself
- Ask each tool for the filing form, date and accession number, then open that filing on EDGAR.
- Compare the "as of" date with the newest filing on the company's EDGAR page; a stale filing was the commonest wrong-figure cause across the tools we tested.
- Add classes of securities, or insiders, yourself rather than trusting one total.
- Run the same question in a fresh chat on both tools before you compare them.
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.
If your question is a filing fact about a US small cap, Signal8 Octo is the tool on this lab built specifically for it. It ranked #1 in both blind rounds: 316 of 320, the same as Claude and 12 above ChatGPT, with no wrong figure. Signal8 makes Octo and runs this lab, so weigh that result with the same caveats as the two above.
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