guides

2026-11-11: 7 API complaints we heard today

PoYo Team
13 min read
Share:

2026-11-11: 7 API complaints we heard today

Complaints that English-speaking developers posted publicly around Nov 10–11, 2026, paraphrased below with links to the original posts. OpenAI's status page said Costs API exports published between Nov 6 and 9 contained incorrect cost data; the incident was identified at 12:42 PM and marked resolved at 3:02 PM Asia/Shanghai on Nov 10 (04:42–07:02 UTC, about 2 hours 20 minutes). Anthropic's status page later recorded four minutes of overloaded errors on Fable 5.2 and Mythos 5.2 (mitigated 5:47 AM Asia/Shanghai / 21:47 UTC on Nov 10). Stripe and OpenRouter also posted that the acquisition had closed. Reddit was quieter for fresh first-person API threads in this window; most of what follows comes from X and official advisories. The louder pain was a cost export you could not trust, a gateway whose payments company now owns the router, a vendor breach that reached billing data and chat transcripts, subscription math that made one Fable token feel like ten Opus tokens, a short Fable/Mythos overload, a 401 that was not "your key is wrong," and video labs that still disagree on physics — and on the bill. Each answer uses only what PoYo's public site and docs say today, and links to the page it comes from. Where the docs don't cover something, we say so.

Q1. The Costs API said one number. Finance may have seen another.

What developers are complaining about: OpenAI reported that Costs API exports published between November 6 and 9 had incorrect cost data. The status page moved from monitoring to resolved in about 2 hours 20 minutes on Nov 10 (identified 12:42 PM, resolved 3:02 PM Asia/Shanghai). Trackers relayed the same window: exports developers use to download what they spent contained wrong figures, and OpenAI said it was correcting the affected records (@petersykim, @steveDOTplus, resolved). If your finance spreadsheet is built on those exports, you had a few days of numbers you should not have booked blindly.

How PoYo answers: PoYo does not document an OpenAI-style Costs API export. What the public docs do give you is a live wallet figure and a retention window you can plan around:

  • Inspect GET /api/user/balance for a real-time credits_amount (user balance). Cache it; do not hammer the endpoint.
  • Chat and media share prepaid credits from one wallet. Credits never expire, and there is no subscription that later invoices "extra usage" (homepage).
  • Empty balance returns 402 insufficient_credits_error, not a surprise invoice (error codes).
  • Failed generation tasks are not charged (homepage, llms.txt).
  • Billing details and generation history are stored for 3 months, then deleted. Download anything you need for your own ledger before that window closes (Getting Started).

Public docs do not describe a formal cost-export API, a guarantee that a third-party invoice will match credits_amount, or a credit adjustment SLA if an upstream provider later revises its own usage file.

Q2. The payments company now owns the model router.

What developers are complaining about: On Nov 10, Stripe posted that OpenRouter is now part of Stripe. OpenRouter's own close post said more than 10 million developers and companies use it, and repeated: "Our commitment to neutrality stays the same. Models and providers compete on quality, reliability, and price. You decide what works for your business" (@OpenRouter). The deal had been announced in August. The worry landed in the replies anyway: when the company that moves the money also routes the tokens, is routing still neutral, or does it become a placement business (@TheGaryMa)? One builder asked whether the neutrality promise covers billing, and whether teams will be able to keep paying outside Stripe long term (@zkToha). Another compared the pledge to past acquisition promises that later changed, and said they were hoping for the best but not feeling good about it (@9K6hf1Kj6B).

How PoYo answers: Always send an explicit model ID and read that card's live schema at https://poyo.ai/models/<id>/llms.txt (llms.txt, API overview). Unknown chat IDs return 404 model_not_found, not a silent remap (error codes). A model only accepts the API formats marked for it; sending the wrong shape returns 400 with the list of supported URIs (chat overview).

Billing stays a prepaid wallet you control: credits never expire, and there is no subscription (homepage). Empty balance is 402, not an automatic top-up (error codes). Cap a key's hourly / daily / total spend and set an IP whitelist on the API key page (Getting Started).

If a published ID is over capacity on your side, pick another documented ID yourself and record which one you sent. Public docs do not claim automatic gateway failover, a weaker-model substitute when the named ID is busy, a Stripe billing integration, or a neutrality SLA about how models are ranked or priced after a third-party acquisition.

Q3. The vendor breach got wider: billing data, datasets, and chat transcripts.

What developers are complaining about: Fireworks' November 9 update went past the October "credentials only" picture. It now says the affected internal application exposed billing information (billing and shipping addresses, invoice and payment history, account balance, and credit usage; payment card numbers were not stored and were not impacted), and that for certain customers a subset of incomplete model weights and datasets was accessed. A third-party cloud provider that hosted Fireworks R&D machines was compromised on Oct 23–24. Notion's advisory, updated Nov 10, says the confirmed dataset included Notion AI chat transcripts used for Insights, affecting fewer than 1% of workspaces; most chats were unattributable (no user or workspace labels, and no tool-call results). Notion says it contacted identified customers, disabled Insights, and stopped serving open-weight models through Fireworks. The advice to affected customers is blunt: treat the data as exposed, and if anyone pasted passwords, API keys, or other credentials into Notion AI during the incident window, rotate them (@RohanBhanotAI, @NobrainerNg, @gzlin).

How PoYo answers: Treat a key as a secret and assume a leaked one will be used — including a key pasted into another product's chat (Getting Started):

  • Never expose keys in frontend code, mobile apps, or public repositories. Don't paste them into a vendor chat, ticket, or "insights" tool either.
  • Create and revoke keys on the API key page. Give each integration its own key so a rotation does not take down everything.
  • Turn on hourly / daily / total caps and an IP whitelist so only approved server IPs can call the API (Getting Started).
  • Send X-End-User-ID if you need to block one caller without rotating the key (end-user identification).
  • A missing or invalid key returns 401 authentication_error (error codes).

Public docs do not claim breach monitoring, automatic key rotation after a third-party incident, or a refund SLA if a stolen key spends credits before you revoke it. Rotate first, then cap what the next key can spend.

Q4. One Fable token felt like ten Opus tokens.

What developers are complaining about: The day after Fable 5.2 launched, the subscription math was still the story. One Ultra/Max user wrote that a Fable token was burning like 10 Opus tokens: last week their Opus use looked like $14k of API value, while this week's entire weekly Fable usage looked like $3k — and they argued the 50% Fable cap plus token weighting made the "same weekly dollars" pitch a lie (@dummerspast39; those dollar figures are from that post, not from Anthropic). A write-up of Theo's Ultra logs made a similar point: about $15,000 a month of Opus 5.5 equivalent versus about $3,000 of Fable 5.2 (@HUNHOfamilyTH). Pro seats were told Fable is included only up to 50% of the weekly limit through Nov 16, then it flips to usage credits at API rates (@Awanioz, @crsnnn). A "Fable reset" still only restores what you spent on Fable; the rest of the weekly bar stays put (@5low_motion). One tester also said Fable 5.2 burned through usage really fast yesterday, then lasted longer after a later fix (@Splash0ooz).

How PoYo answers: Treat the live catalog as source of truth. Today's chat table lists claude-fable-5-2 (chat overview, llms.txt). Live rates on that card are 1280 credits ($6.40) input and 6400 credits ($32.00) output per 1M tokens, plus the documented cache-read and cache-write rows (claude-fable-5-2/llms.txt, Claude Fable 5.2). Call it on Claude Messages /v1/messages or Chat Completions /v1/chat/completions as marked (chat overview). Authenticate with Authorization: Bearer; /v1/messages also accepts x-api-key (Getting Started).

There is no weekly bar, no 50% model cap, and no "Fable-only reset" that leaves the rest of the wallet locked: prepaid credits live in one wallet and never expire (homepage). Poll GET /api/user/balance and stop on a hard 402 (user balance, error codes). Cap keys so an idle agent cannot keep spending (API key page).

Public docs do not offer subscription-style weekly limits, launch-week Pro windows, or a promise that a seat reset will restore other models' usage.

Q5. Fable and Mythos had a four-minute overload.

What developers are complaining about: Anthropic's status page recorded elevated errors on Fable 5.2 and Mythos 5.2, then marked the incident resolved. The note says overloaded errors lasted four minutes, mitigated at 5:47 AM Asia/Shanghai on Nov 11 (21:47 UTC on Nov 10). Short windows still break a coding agent mid-turn if your retry loop does not know a 529/overloaded from a bad request.

How PoYo answers: Getting Started already sets a realistic stability expectation: overall stability may be slightly lower than official providers. When something fails, read the documented code instead of hammering retry:

  • 500 / 502 / 503 — retry with exponential backoff. 502 is an upstream_error (error codes).
  • 429 rate_limit_error — wait or queue; do not treat it as "add money." LLM endpoints are currently not rate limited (chat overview).
  • 402 insufficient_credits_error — empty wallet; top up (error codes).

If you want a different documented chat lab while an upstream window is noisy, send that lab's published ID — for example Claude Opus 5.5, Claude Sonnet 5.5, or Gemini 4 Argon (chat overview). Public docs do not claim automatic multi-lab failover, list a mythos model ID, or promise that PoYo stayed up when Anthropic did not. An unknown chat ID returns 404 model_not_found (error codes).

Q6. The 401 was not "you pasted the wrong key."

What developers are complaining about: Cloudflare posted a Replicate incident, Prediction issue on OpenAI models via Replicate: some predictions on official OpenAI models failed with a 401 token_invalidated error. The window ran from 2:35 AM to 3:06 AM Asia/Shanghai on Nov 11 (18:35–19:06 UTC on Nov 10, about 31 minutes) and was marked resolved. Status trackers relayed the same failure (@outageslive). A 401 that appears mid-job is easy to misread as a revoked key or a bad secret in your env.

How PoYo answers: On PoYo, HTTP 401 authentication_error means the API key is missing, malformed, invalid, or deleted — check Authorization: Bearer (or x-api-key on /v1/messages) and whether the key still exists (error codes, Getting Started). That is a different event from an upstream token_invalidated on another gateway.

Generation uses the unified async path: POST /api/generate/submit → status poll or webhook (API overview, webhooks). Failed generation tasks are not charged (homepage, llms.txt). Do not blindly resubmit: a retry can create another billable task if the first already accepted (Getting Started). For transient 500 / 502 / 503, back off; 502 is an upstream_error.

Public docs do not describe a token_invalidated 401, Replicate compatibility, or a refund for a failed prediction on another platform. If you see 401 on PoYo, fix the key first; do not retry it as if it were a 503.

Q7. The same 10-second action prompt still produces three different clips — and three different bills.

What developers are complaining about: Quality and price still diverge once you leave the landing page. One builder ran Gemini Omni 1.2 Flash, Seedance 2.5, and Kling 4.0 on the same 10-second multi-cut action prompt (zipline escape + hatchet cut) and reported a spatial-logic miss on Omni (the hero slid toward the pursuers), strong vector awareness on Seedance 2.5, and the cleanest multi-cut on Kling 4.0 (@NextGenAIVoult). That bake-off is from their run, not from PoYo — but it is the usual production question: which documented ID do you actually submit, and what will that ID cost if the first take is wrong?

How PoYo answers: Video uses the unified async path: POST /api/generate/submit → status poll or webhook (video overview, webhooks). Failed generation tasks are not charged (homepage, llms.txt). Send an explicit ID and read live per-second prices on the card you will actually call (re-check the file before you budget). For a 10-second clip at 720p with no video input, today's cards are:

Ops tips:

  • A cheaper documented ID is the right tool for iteration; a keeper ID is a separate submit.
  • Prefer webhooks over aggressive polling.
  • Do not blindly resubmit: a retry can create another billable task if the first already accepted (Getting Started).
  • Download finished files within 3 days retention (Getting Started).
  • Cap spend with key limits and stop on 402 (API key page, error codes).

Docs do not promise a dislike / regenerate refund for finished-but-undesired clips, automatic multi-lab video failover, or that another site's physics result will match a PoYo submit of the same ID.

Quick map

ComplaintWhere PoYo documents it
OpenAI Costs API exports with wrong numbersBalance API · prepaid wallet · credits never expire · no subscription · 402 · failed tasks not charged · 3-month billing history · no cost-export API / invoice SLA
Stripe closes OpenRouter / neutrality and billing questionsExplicit IDs · live llms.txt · 404 not a silent remap · prepaid credits · key caps · no auto-failover or Stripe billing integration
Fireworks / Notion wider disclosure (billing data, transcripts, pasted keys)Secret handling · rotate on API key page · caps + IP whitelist · X-End-User-ID · 401 · no breach-monitor / refund SLA
Fable 5.2 quota math / 50% caps / reset theaterCatalog lists claude-fable-5-2 · live 5.2 llms.txt · Messages or Chat Completions · one wallet · no weekly bar or 50% cap
Four-minute Fable / Mythos overloadStability note · 500/502/503 backoff · 502 upstream_error · no auto-failover · unknown IDs are 404
Replicate 401 token_invalidated on official OpenAI modelsPoYo 401 is a bad/missing key · failed tasks not charged · no blind resubmit · 502 upstream_error · no token_invalidated code
10-second Omni / Kling / Seedance bake-offPer-second live cards · failed tasks not charged · webhooks · 3-day retention · no dislike refund

Browse models and live prices on poyo.ai, schemas via llms.txt, and the full docs index at docs.poyo.ai/llms.txt.

Share: