Total — unique visitorsRun by AI agents · Live
Browse by category General AI Assistants AI Agents Writing and Translation AI Search and Research Image Generation Video Generation Voice and Music Marketing and SEO Coding Productivity AI Aggregators and APIs App and Site Builders Free Tier Table

AskCat Daily: PRICE HIKE, TIER CUT — we'll warn you first

AMPM STANDARD

Transparencies

Scope of the data: 288 monitored sources, 273 published tools, a difference of 15. The 15 not published are monitored items that have not met the publication bar: Sora (discontinued by its vendor; kept on record but not listed as a public tool), Moltbook (not yet eligible for publication), OpusClip (not yet eligible for publication), Echo Loop (not yet eligible for publication), Runable (data not yet complete, hidden until it is), whacka (data not yet complete, hidden until it is), Ranked AI (data not yet complete, hidden until it is), Chipp (data not yet complete, hidden until it is), Viktor (data not yet complete, hidden until it is), Subscribr (data not yet complete, hidden until it is), Heyy (data not yet complete, hidden until it is), Magica (data not yet complete, hidden until it is), Fyxer (data not yet complete, hidden until it is), Fabric (data not yet complete, hidden until it is), Weav (data not yet complete, hidden until it is).

Rankings can be questioned, so the methodology must be transparent. This page explains how AMPM100 is selected, how scores are calculated, how evidence is verified, and how money and rankings are separated.

AI at work, liveWhat the AIs behind this site are doing right now, where something went wrong and how it was handled. Updated every 5 seconds from real server records.Open the live view →

1. How AMPM100 is selected

AMPM does not list every AI tool; it only continuously monitors the 100 most noteworthy. For a tool to become an official member (Confirmed Active), it must satisfy the following definition:

Confirmed Active is not equivalent to a computable score. It must fully pass the eight Active eligibility checks of AMPM scoring standard: evidence coverage>=80%, zero unresolved conflict, >=4/5 dimensions OK, primary category confirmed, identity/product status confirmed, pricing/free evidence valid, confidence/evidence gate passed (requires critical facts to complete a genuine two-pass verification, not merely legacy transfer), lifecycle state=ACTIVE_ELIGIBLE. Any failure results in Under Review, and standards must not be lowered to meet quotas.

Items that fail should always display Under Review and “Pending verification,” without padding with fake scores.

2. How AMPM Score is calculated

The AMPM Score is Wenwen Cat’s overall rating for each tool (0–100). The calculation method is disclosed below, but the itemized internal numbers for each tool are not disclosed (to prevent manipulation).

  • Worth the money?: 25%
  • Reliability: 20%
  • Ease of use: 15%
  • Practicality: 30%
  • Growth: 10%

Methodology Version: ampm-score-v2.1.0

3. How the evidence is verified

Each publicly disclosed price and fact must be traceable to its official source and capture time. Key facts (price, free quota, plan changes) require two independent technical pipelines to each obtain consistent results (two-pass) before verification is considered complete; when only a single source is available, the confidence level must be indicated truthfully, insufficient data is shown as “pending verification,” and we never fill in 0, guess, or use model memory as fact.

4. Recheck frequency

The system regularly checks every monitored source automatically: a standard fetch compares the official pages, and sources that cannot be fetched that way or are blocked are re-fetched with a headless browser (both pipelines are change detection with zero AI cost; they only tell whether something changed and do not parse amounts). The master database is then rebuilt, so the whole site shares one data snapshot. A detected change is only written as an official change after it shows up again after an interval, so promotional flickers and A/B tests do not create false events. Official prices are also checked by the automatic price corrector: it reads the official page on two different days (this step uses AI to extract plan prices from the page text), updates the price and stamps the check date only when both reads agree, and flags a mismatch for a human.

Data current as of: 2026-11-20T03:25:11+08:00 | Tools monitored: 288

Verification dates for 273 live tools: 220 within 3 days, 9 within 4-7 days, 44 older than 7 days.

Column pipeline: 50 new articles in the last 7 days (1 today). Latest: [Pricing] Clico Worth It? What to Pin Down Before You Subscribe (2026-11-20)

Loop relay (computed at each build): Sense: 8 news items ingested in the last 24 h → Decide: 6 price changes confirmed in 2026-10 (4 tools) → Build: 50 new articles in the last 7 days → Check: 220 of 273 tools verified within 3 days (plus 0 re-checked by a human in a real browser within 3 days) → Link: 951 article pages (all language versions) bound to verified prices, flagged automatically when a price changes.

Loop-back (topic-review cycle, snapshot 2026-11-20 00:52 Taipei time): The topic-review cycle keeps running (last lap 11-20 00:50). Totals: 195 events taken in, 312 expired or duplicate working-memory items tidied, 48 items promoted to long-term knowledge, 0 evolution proposals (0 means none, never padded). This station writes each lap's result back to memory, and the next lap's sensing and decisions start from there.

Script self-check (computed at each build): The automation runs 161 programs; each is statically checked (read only, never executed) for undefined names that only crash when that line is reached. Programs with a problem right now: 0. A price-correction job crashed this way on 2026-10-14; it was fixed and re-run successfully the same day, and this check then joined the regular patrol.

Price-check red flags and human review: 19 tools currently have official-site prices that the automatic reader cannot match (usually a login wall, price cards that do not render, euro or NT dollar display, or API-only pricing), so a machine cannot decide. Each one is compared by AM&PM (the maintenance agent) against real-browser snapshots and written into the check log of that tool; 0 of them are unchanged since review and are not re-raised for 14 days, and 19 are still lit and waiting. If the reason for a red flag changes it lights again at once, and after 14 days it is forced back for review; site prices are never changed automatically until two official reads agree.

AMPM100 reconciliation: The AMPM100 data files are compared number by number with 5 pages (home, AMPM100, challengers, and the English home and AMPM100 pages). All 12 metrics agree right now (checked 11-20 00:53). This check used to run only when someone asked; one probation count once used a different definition for more than a day unnoticed. Patrol now runs it hourly, and any mismatch goes into the patrol list and the loop ledger; the figures on this page refresh with the public snapshot every 3 hours (check time in brackets).

Hot topics and similarity gate (computed at each build): 2 hot topics suited to readers today (0 written, 2 waiting, 0 short on sources): Gemini (7), Gemini (6). When several outlets cover the same event separately, they are merged into one story and their sources are added up; a hot topic with enough sources is written before evergreen pieces. A new draft whose content is 30% or more similar to an existing article on the same tool is sent back to be rewritten: 0 rejections today (latest similarity -%). If three rewrites are still too close, it is not published.

Is external intelligence being used? (computed at each build): 1838 external items absorbed so far, indexed by tool (21 tools covered, kept for 7 days). Uses: fed to the writer as background with a source-citation requirement (196 items offered, 15 actually cited in published articles); tools discussed by 2+ different external sources in the last 3 days get a topic-selection bonus (1 times applied); 65 price-related items sent to price re-checks (4 tools also mismatch on the official-site check); gaps where outsiders are talking but this site has not written in 14 days: Elevenlabs (2), Perplexity (2), Heygen (2), Suno (2).

Scorecard (last 7 days, computed at each build): Scheduled jobs ran 49501 times with 281 failures, a 99.4% success rate; deploy batches: 38 shipped, 6 stopped by the gate. Failures and stops are listed as they are; a stop usually means the gate caught a problem before it went live, not that the site broke.

Deploy deferrals (last 7 days): 39 deploy flags were deferred in the last 7 days (environment-only pre-check 22, waiting for a work ticket 15, template half-edited 2). A deferral means yielding on purpose, not failing: the flag stays queued and ships in the next deploy slot, nothing is dropped. Since 14 Oct, if the pipeline is busy right after pre-checks it now waits up to 6 minutes before deciding.

Day by day
  • 2026-11-14: 5584 job runs, 47 failed (99.2% success); 5 deploy batches shipped, 4 stopped
  • 2026-11-15: 5604 job runs, 63 failed (98.9% success); 0 deploy batches shipped, 1 stopped
  • 2026-11-16: 8810 job runs, 40 failed (99.5% success); 4 deploy batches shipped, 1 stopped
  • 2026-11-17: 9733 job runs, 19 failed (99.8% success); 12 deploy batches shipped, 0 stopped
  • 2026-11-18: 9743 job runs, 89 failed (99.1% success); 3 deploy batches shipped, 0 stopped
  • 2026-11-19: 9668 job runs, 21 failed (99.8% success); 14 deploy batches shipped, 0 stopped
  • 2026-11-20: 359 job runs, 2 failed (99.4% success); 0 deploy batches shipped, 0 stopped

Links between modules (tested at every build): The registry lists 22 “who hands information to whom” links; 21 are working and 1 are not (Controller→Evolution). Each link is judged by whether the file is fresh and the receiving program really reads it — not a hard-coded green light. Added this round: Controller→Evolution (the deploy ledger written for every release is read by the watchdog and the public scoreboard). If any link stays broken for more than 2 hours, the controller automatically files a repair ticket.

Show all links

Scheduler→Sensor, Scheduler→Controller, Sensor→Controller, Controller→DataBus, DataBus→Controller, DataBus→Chassis, Chassis→Casing, Controller→Production, Production→Chassis, Monitor→Scheduler, Logbook→Controller, Interlock→Scheduler, Monitor→Cleanup, Monitor→Console, Engine→Logbook, Sensor→Logbook, Controller→Repair, Inspection→Controller, Repair→Monitor, Evolution→Inspection, Production→Inspection, Controller→Evolution

Safeguards on automatic price edits (last 7 days, recomputed at every build): An automatic price edit happens only when two official-site reads agree. Last 7 days: 5 automatic edits, 1 of them later found wrong and reverted (instapage); 0 prices that appear only after “Then / renews at” (a discounted renewal rate, not the list price) were rejected; 62 checks confirmed the listed price unchanged. Incident 10-17 (kling): the script took a first-renewal discount for the list price, which made the price hard-gate block deployments for about 1.5 hours; the tool page briefly showed a “price being confirmed” notice containing those wrong numbers, now withdrawn, and the listed prices were never changed. Two safeguards were added: renewal-only prices are rejected, and every automatic edit is re-checked against the hard gate and rolled back if it fails.

Repair loop (last 7 days, recomputed at every build): 114 scheduled-job failure incidents (consecutive failures of one job count as one); 111 recovered, 3 still open. Time to recover: median 5 min, 79% within 30 min, 90% within 536.5 min, longest 3459.9 min. 93 recovered on the very next run, 21 failed 2+ times in a row; 5 jobs had 2+ incidents this week (recurrence tracking, root cause is re-checked). A job that runs once a day cannot recover faster than the next day, so the longest figure is mostly schedule spacing, not slow repair.

Incidents per day (MM-DD count): 11-14 24, 11-15 8, 11-16 9, 11-17 6, 11-18 37, 11-19 4, 11-20 1

Real-visitor error radar (Cloudflare, whole domain incl. the main site): In the last 23 hours there were 202724 requests. Visitors hit 0 server errors (5xx) and 2955 not-found responses (404/410, scanners excluded), an effective error rate of 1.46%. A further 14617 failures came from the CDN own prefetch sub-requests, which are not visitors and are counted separately. Our own server log (sub-sites) also showed 10 search/AI-crawler hits on 404 pages in 24 hours; those are being turned into 410 or relinked. Counts only, no URLs.

Inspection rejection funnel (last 7 days, recomputed at every build): Over the last 7 days the main site wrote 67 Chinese articles; 60 went straight live (90%) and 7 were held back as drafts. Today: 1 written, 1 live, 0 held. Held does not mean the writer idled: the inspection layer (automatic review and manual recall) is doing its job.

Why they were held: 0 duplicate or too similar to a published piece, 0 numbers with no traceable source, 4 failed other review checks, 3 manually recalled (off-topic etc.), 0 default drafts waiting for review.

Live / written per day: 11-14 3/7, 11-15 7/8, 11-16 7/7, 11-17 14/15, 11-18 15/16, 11-19 13/13, 11-20 1/1

Sub-site auto-deploy paused for a human look: Each sub-site has an auto-deploy job. Paused right now: 0. A pause is a safety feature, not a breakage: whenever a sub-site has template or config edits nobody has reviewed yet, the job stops and waits for a person, so half-finished work never goes live by itself.

How long a written article takes to go live: In the last 48 hours the writer produced articles in 12 shifts; 12 are live (0 still waiting). From the end of a shift to the first successful deploy the median is 0 minutes and the longest is 83 minutes; 9 shifts deployed by themselves right after writing (0 minutes). Latest shifts (end time: minutes): 11-19 18:09: 0, 11-19 22:15: 0, 11-19 23:42: 0, 11-20 00:47: 0. The slow ones were mostly held back because other windows were editing the site (the writer does not deploy over unregistered changes) or because price-monitoring noise closed the gate.

Deploy-gate monitoring quality (avoids blocking the whole site by mistake): In the latest price-check round (2026-11-19 16:08 UTC) we watched 0 official sources; 0 failed over plain HTTP: 0 are on the registered long-term blocked list, 0 were rate-limited (HTTP 429) but read fine by the browser layer within the last 4 hours () so they are not blind spots, and 0 were truly unread. The gate is open. Previously one rate-limited source blocked every deploy on the site (example this afternoon: plaud returned 429 and the 16:00 pre-deploy check refused to deploy).

Pre-flight check before a deploy flag is queued (avoids collateral blocks): In the last 7 days 4 deploy batches were stopped by the main gate, involving 6 flags in total (innocent flags in the same batch get blocked together). Before a flag is queued we now do a one-minute trial build and scan the English pages for stray Chinese text; a bad flag is turned back on the spot and nothing is queued. Last 7 days: 14 checks, 14 passed, 0 stopped. Latest: 2026-11-16 14:53.

Derived-page date catch-up (cat reviews, free-tier details, plan details, comparisons, deals; verified on a regular schedule every day): Of the 915 reader-visible derived entries, 846 already carry the source tool's latest check date; 18 are in the rewrite queue because the prices in the text no longer match the current list price (old date kept, not stamped); 44 wait for the next shift because the source tool was just re-checked; 0 should have been stamped but were not. The last verification shift ran at 2026-11-19 22:10.

Writer shifts that deploy on their own (last 7 days, computed at every build): The writer ran 17 shifts. After writing, it first checks whether anyone else has unregistered changes in the source; if so it does not deploy itself and the article waits for the next flag batch. 3 shifts were held back (today 0) and 0 failed a gate. Per day (held/shifts): 11-19 3/16, 11-20 0/1. Most holds were scheduled jobs own outputs (snapshots, reports) mistaken for someone else edits; those files are now listed as routine output.

SEO title experiments (computed at every build): To match what people actually search, we have auto-edited the titles of 166 pages (0 today). Each edit is re-checked against Search Console data after 7 days and automatically rolled back if impressions fall too far (8 rolled back so far). Checking each content file: 110 pages still carry the new title; 48 were later rewritten or wiped, so they do not count as valid experiments: /en/tools/felo/, /en/guides/felo-free-usage-limits-2026/, /en/guides/ai-subscription-cancel-refund-app-store-2026/, /en/guides/claudeforce-salesforce-open-beta-2026/, /en/guides/claude-fable-5-1-copilot-data-retention-2026/, /en/compare/claude-vs-lovable/, /en/tools/pika/, /en/guides/copilot-4-models-deprecated-oct-2-2026/, /en/guides/deepl-glossary-zh-hant-v3-setup-2026/, /en/guides/grok-pricing-2026/, /free/, /en/guides/leonardo-howto-2026/. 45 pages are waiting for their 7-day review, the earliest on 2026-11-20.

Price-sensor coverage (computed at every build): The auto price check runs in two layers: plain page reads watch 349 official sources (328 read successfully in the last 24 hours; 21 block automated reads with HTTP 4xx, so they are backed off and retried once a day), and browser reads watch 156 more (151 succeeded in the last 24 hours). Of 269 tools, 267 had a readable page in the last 48 hours and the price extractor pulled a price for 243. The other 24 are covered by a separate daily check plus manual review; their latest record is: 8 fully matched by the daily check, 3 manually reviewed, 13 mismatched (or have no list price but showed numbers) and are waiting for a human to confirm (check date is not advanced until then). List: adobe-firefly, aider, canva-ai, capcut-ai, clipchamp, comet, createos-sandbox, gamma, google-ai-studio, ideogram, landr, leonardo, livedocs, luma, magic-hour, mistral, moltbook, monica, poe, remove-bg, sora, trae, vibe-trading, yylo. Last full sweep: 2026-11-19 16:06 UTC.

How the price sensor handles blind spots: Some official pricing pages show only one tab by default and hide other plans behind tabs. We now register those pages and click the tabs open before reading prices: 1 site(s) (leonardo.ai (team)); for example Leonardo's Team plans used to be invisible to the reader and are now read and match our listing. 1 site(s) show prices only after login (hailuo-ai); the robot sees only an entry price, so these are marked for manual price checks instead of guessing from the home page.

Module pulse (17 modules: 13 on time (0 of them near the stall limit), 4 event-driven (idle when nothing happens), 0 overdue)
  • Scheduler: running, last activity 0.1 h ago (stall limit 0.5 h)
  • Engine: running, last activity 0.1 h ago (stall limit 0.5 h)
  • Sensor: running, last activity 2.5 h ago (stall limit 8 h)
  • Controller: running, last activity 2.5 h ago (stall limit 8 h)
  • Data bus: running, last activity 2.7 h ago (stall limit 30 h)
  • Logbook: running, last activity 0.5 h ago (stall limit 7 h)
  • Interlock: running, last activity 1.0 h ago (stall limit 26 h)
  • Inspection: running, last activity 0.3 h ago (stall limit 4 h)
  • Monitor: running, last activity 0.1 h ago (stall limit 0.5 h)
  • Repair: event-driven (no fixed schedule)
  • Production: running, last activity 0.0 h ago (stall limit 3 h)
  • Console: running, last activity 0.0 h ago (stall limit 0.5 h)
  • Cleanup: running, last activity 20.5 h ago (stall limit 30 h)
  • Evolution: running, last activity 2.5 h ago (stall limit 26 h)
  • Assembly: event-driven (no fixed schedule)
  • Agent: event-driven (no fixed schedule)
  • Controlled rebuild: event-driven (no fixed schedule)

Machine-readable (CC BY-SA 4.0): /data/loop-status.json

Older ones are mostly cases where the official page was redesigned, shows region-based prices or bills by usage, so automated checks cannot match them line by line. We do not stamp a fresh date on them; the old date stays until a person re-checks, so every date you see is real. The list below is ordered by search demand: tools people look up most are re-checked first.

Tools last verified more than 7 days ago (44)
  • Perplexity: 2026-09-08 (72 days ago) · page was read but no Taiwan price could be matched, not stamped · affects 13 compare pages
  • Google AI Studio: 2026-10-10 (40 days ago) · page was read but no Taiwan price could be matched, not stamped · affects 19 compare pages
  • Manus: 2026-10-30 (20 days ago) · page was read but no Taiwan price could be matched, not stamped · affects 7 compare pages
  • VEED: 2026-10-20 (30 days ago) · page was read but no Taiwan price could be matched, not stamped
  • Gamma: 2026-11-04 (15 days ago) · page was read but no Taiwan price could be matched, not stamped · affects 6 compare pages
  • TRAE: 2026-11-08 (11 days ago) · page was read but no Taiwan price could be matched, not stamped
  • PixVerse: 2026-10-03 (47 days ago) · page was read but no Taiwan price could be matched, not stamped
  • DeepL: 2026-10-03 (47 days ago) · page was read but no Taiwan price could be matched, not stamped · affects 2 compare pages
  • Adobe Firefly: 2026-10-27 (23 days ago) · page was read but no Taiwan price could be matched, not stamped · affects 7 compare pages
  • aider: 2026-10-10 (40 days ago) · page was read but no Taiwan price could be matched, not stamped · affects 7 compare pages
  • Speechify: 2026-10-05 (45 days ago) · page was read but no Taiwan price could be matched, not stamped
  • Lovable: 2026-10-03 (47 days ago) · page was read but no Taiwan price could be matched, not stamped · affects 6 compare pages
  • Context.dev: 2026-10-25 (25 days ago) · page was read but no Taiwan price could be matched, not stamped · affects 6 compare pages
  • Stable Diffusion: 2026-10-26 (24 days ago) · page was read but no Taiwan price could be matched, not stamped · affects 7 compare pages
  • QuillBot: 2026-09-08 (72 days ago) · page was read but no Taiwan price could be matched, not stamped
  • Hailuo AI: 2026-10-09 (41 days ago) · page was read but no Taiwan price could be matched, not stamped
  • Poe: 2026-10-11 (39 days ago) · page was read but no Taiwan price could be matched, not stamped · affects 10 compare pages
  • Kagi: 2026-10-23 (27 days ago) · page was read but no Taiwan price could be matched, not stamped
  • SmugMug: 2026-11-03 (16 days ago) · page was read but no Taiwan price could be matched, not stamped
  • YYLO: 2026-09-27 (53 days ago) · page was read but no Taiwan price could be matched, not stamped
  • Murf AI: 2026-09-24 (56 days ago) · page was read but no Taiwan price could be matched, not stamped
  • Rytr: 2026-10-16 (34 days ago) · page was read but no Taiwan price could be matched, not stamped
  • DeepInfra: 2026-10-23 (27 days ago) · page was read but no Taiwan price could be matched, not stamped
  • Landr: 2026-10-29 (21 days ago) · page was read but no Taiwan price could be matched, not stamped
  • Postman: 2026-11-01 (18 days ago) · page was read but no Taiwan price could be matched, not stamped
  • Campaign Monitor: 2026-10-23 (27 days ago) · page was read but no Taiwan price could be matched, not stamped
  • Klaviyo: 2026-10-21 (29 days ago) · page was read but no Taiwan price could be matched, not stamped
  • AIVA: 2026-10-23 (27 days ago) · page was read but no Taiwan price could be matched, not stamped
  • Comet: 2026-09-08 (72 days ago) · page was read but no Taiwan price could be matched, not stamped · affects 9 compare pages
  • Rev AI: 2026-10-24 (26 days ago) · page was read but no Taiwan price could be matched, not stamped
  • Nitro PDF: 2026-10-23 (27 days ago) · page was read but no Taiwan price could be matched, not stamped
  • You.com: 2026-10-24 (26 days ago) · page was read but no Taiwan price could be matched, not stamped
  • Travis CI: 2026-10-25 (25 days ago) · page was read but no Taiwan price could be matched, not stamped
  • Missive: 2026-11-02 (17 days ago) · page was read but no Taiwan price could be matched, not stamped
  • Hiver: 2026-11-11 (8 days ago) · page was read but no Taiwan price could be matched, not stamped
  • SandBase CLI: 2026-10-22 (28 days ago) · page was read but no Taiwan price could be matched, not stamped
  • Together AI: 2026-10-26 (24 days ago) · page was read but no Taiwan price could be matched, not stamped
  • PostHog: 2026-10-28 (22 days ago) · page was read but no Taiwan price could be matched, not stamped
  • Mubert: 2026-11-04 (15 days ago) · page was read but no Taiwan price could be matched, not stamped
  • CreateOS Sandbox: 2026-11-05 (14 days ago) · page was read but no Taiwan price could be matched, not stamped · affects 3 compare pages
  • Clico: 2026-11-05 (14 days ago)
  • Circleback: 2026-11-11 (8 days ago)
  • VoiceOS: 2026-11-11 (8 days ago)
  • Appy.AI: 2026-11-11 (8 days ago)

5. How to correct errors

When we find an error, we correct the data itself and sync every place it is cited; the correction log is public at /audit/.

6. Commercial firewall

AMPM has affiliate links and ads, but they live only in the final monetization layer and are labelled "Ad" on the page: affiliates, sponsors and ads do not affect the AMPM Score, category rank, seats, ENTER/DROP, Battle results or Ask answers, and they do not affect museum inclusion either. Three things stay separate: ranking reflects only the score; the museum admits tools only on public historical criteria, with or without a partnership, and the museum page carries no recommendation or ad cards; affiliate links and ads appear only in labelled commercial blocks. Ranking first, monetization second. Affiliate links are always labelled and go through /go/ redirects.

Affiliate Links Notice

7. Data file version

coverage: 2026-11-18T21:00:03Z
eligibility: 2026-11-19T19:25:04Z
radar: 2026-11-19T16:52:03Z
score: 2026-11-19T19:25:11Z
taxonomy: 2026-08-19T08:00:00Z
tools: 2026-11-20T01:28:19+08:00

8. How the system runs

How AMPM Works: An Automated Site Built From 15 Modules: what each of the modules does, which run today and which are unfinished.

Advertisement