DeepSeek представит последнюю версию v4

DeepSeek выпустит v4 в середине июля с новой моделью ценообразования API: токены подорожают в 2 раза в часы пиковой нагрузки (09:00–12:00 и 14:00–18:00 по пекинскому времени). Компания планирует уведомлять пользователей по почте за 24 часа до изменения тарифов. Проблема с ошибками «server busy» останется, но обойдётся дороже — это может существенно повлиять на экономику проектов, которые активно используют API DeepSeek для автоматизации и масшта…

➡️ Читайте на сайте: https://aff.top/blog/deepseek-predstavit-posledniuiu-versiiu-v4

🧠 Ещё больше инсайтов → в канале AFF.top

Why frequency caps leak across the open programmatic graph

A frequency cap (limit how many times one user sees your ad) sounds like a counter. In open programmatic it's closer to an estimate, and the leak is structural, not a bug.

1. Capping requires stable identity. When match rate is partial and identifiers fragment across exchanges, the same person can appear as several distinct users to your DSP.
2. Each apparent identity gets its own cap allowance, so a user who resolves to three IDs can see your ad three times the intended limit.
3. Cross-device and in-app environments amplify this; the deterministic link between a phone and a desktop is often missing.

The consequence is overexposure on your heaviest users and wasted impressions that also depress performance through ad fatigue — the opposite of what the cap was meant to prevent.

Mitigation leans on a unified identity layer and, where deterministic IDs are absent, probabilistic stitching with known error rates. But you should treat your cap as a soft target with a leakage percentage, not a hard ceiling.

Why it matters: if you report a 3x cap but your true distribution has a tail of users seeing your ad 8–10 times, your reach and effectiveness numbers are both wrong. Measure realized frequency from log-level impression data against your set cap; the gap is your identity fragmentation, quantified.

Anthropic выпустили Sonnet 5

30 июня вышла Claude Sonnet 5 — новая версия позиционируется как самая агентная в линейке и приближается к флагманской Opus 4.8. Модель лучше справляется со сложными многоуровневыми задачами, устойчива к вредоносным запросам и не генерирует эксплойты. Sonnet 5 доступна на Free-тарифе, но тестирование показало скромные улучшения: хотя работает лучше Sonnet 4.6, её обгоняют конкуренты, включая китайские модели, которые дешевле через API при лучшей…

➡️ Читайте на сайте: https://aff.top/blog/anthropic-vypustili-sonnet-5

🧠 Ещё больше инсайтов → в канале AFF.top

Clickstar прекращает работу

Clickstar закрывается. Легендарная пуш-сеть прекращает закуп трафика с 1 августа, полная остановка — 20 августа.

Сетка работала почти 8 лет и была одним из лучших источников качественного трафика на Россию и СНГ. Сейчас пуш-трафик стал слишком ботовым из-за гугловских банов на скрипты сбора.

Что это означает для арбитражников — разбираемся в ста…

➡️ Читайте на сайте: https://aff.top/blog/clickstar-prekraschaet-rabotu

🧠 Ещё больше инсайтов → в канале AFF.top

Auction density and why pacing is harder than it looks

Budget pacing spends evenly over a flight instead of front-loading. The naive model assumes auction opportunities arrive uniformly. They don't, and the non-uniformity breaks simple pacing.

1. Auction density — the rate of eligible auctions for your target — varies by hour, day, and audience, sometimes 5–10x between trough and peak.
2. A pacing system that targets constant spend-per-hour will either starve during low-density hours (no inventory to buy) or get forced to overpay to hit its hourly number.
3. Worse, raising bids to hit a pacing target during a low-density window pollutes your clearing-price data, teaching the shading model that inventory clears higher than it really does.

The better model paces against available opportunity, spending more when good inventory is abundant and cheap and less when it's scarce and expensive — which usually means accepting an uneven hourly spend curve to get an even-quality outcome.

Why it matters: pacing and bidding are coupled, not independent. A pacing rule that forces spend during scarce hours quietly inflates your CPMs and corrupts the very model that's supposed to keep prices honest. Evaluate pacing on cost-per-outcome across the flight, not on how smooth the hourly spend line looks.

Facebook запретил рекламу онлайн-казино Mr Vegas

Британский ASA запретил рекламу казино Mr Vegas из-за «слишком милых» мультяшных животных в креативах — регулятор счёл, что такой стиль привлекает детей, в том числе через Facebook. Рекламодатель запустил кампанию в феврале, бан вышел в июле. Логика регулятора вызывает вопросы: дети неплатёжеспособны, а таргетировать их на гемблинг бессмысленно.

➡️ Читайте на сайте: https://aff.top/blog/facebook-zapretil-reklamu-onlain-kazino-mr-vegas

🧠 Ещё больше инсайтов → в канале AFF.top

В Whatsapp скамят пользователей с помощью поддельных никнеймов

WhatsApp запустил никнеймы — и почти сразу начался скам. Мошенники регистрируют имена, похожие на бренды, звёзд и политиков, с минимальными опечатками.

Индия, где 500 млн пользователей WhatsApp, потребовала от Meta объяснений за 3 дня. Meta говорит, что точные совпадения заблокированы — но одна буква в другом месте защиту не триггерит.

Похоже, п…

➡️ Читайте на сайте: https://aff.top/blog/v-whatsapp-skamiat-polzovatelei-s-pomoschiu-poddelnykh-nikneimov

🧠 Ещё больше инсайтов → в канале AFF.top

Вышел ZCode - аналог Claude code

Вышел ZCode — десктопный аналог Claude Code от разработчиков GLM-5.2. Работает с API от Anthropic, поддерживает SSH-деплой на сервер, в том числе Linux.

Вместо пошаговых скриптов — система целеполагания Goal: закидываешь сложный промт, агент сам разбивает задачу и выполняет. Плюс управление через Telegram-бота.

Но главная фича — мультиагентность…

➡️ Читайте на сайте: https://aff.top/blog/vyshel-zcode-analog-claude-code

🧠 Ещё больше инсайтов → в канале AFF.top

Bid reduction inside the SSP: the cut you can't see from the buy side

Buyers obsess over their own DSP's bid shading, but the sell-side exchange can also alter prices, and that layer is largely invisible from the buy side. Understanding it explains otherwise baffling clearing data.

1. Some SSPs apply their own optimizations — soft floors, dynamic floors, or bid manipulation — between receiving your bid and clearing the auction.
2. Historically, undisclosed sell-side fees and bid caching created gaps where the price you paid and the price the publisher received diverged by more than any declared take rate.
3. Header bidding and the push for transparency (and audits like the ISBA/PwC study that famously couldn't trace ~15% of spend) exist precisely because this layer was opaque.

From your seat, the symptom is a clearing price that doesn't reconcile with the publisher's reported payout for the same impression, beyond the declared margin.

Why it matters: when buy-side and sell-side logs don't reconcile, the unexplained gap lives in this sell-side layer. The defense is the same as everywhere else — favor short, transparent supply paths to publishers who share log-level data, because every intermediary you remove is one fewer place a price can be quietly adjusted between your bid and the publisher's revenue.

Cloudeflare грозит Google блокировкой трафика

Cloudflare объявил: с 15 сентября 2026 года ИИ-краулеры будут заблокированы по умолчанию на всех сайтах с рекламой — включая Googlebot, Applebot и Bingbot.

Главная претензия — к Google: один и тот же бот индексирует страницы и собирает данные для обучения нейросетей, что даёт поисковику нечестное преимущество.

Но есть нюанс, который меняет всю к…

➡️ Читайте на сайте: https://aff.top/blog/cloudeflare-grozit-google-blokirovkoi-trafika

🧠 Ещё больше инсайтов → в канале AFF.top

Гайд: как заработать первые деньги на Pornhub

Pornhub — самый посещаемый адалт-сайт в мире, и на нём действительно можно зарабатывать. Но схема устроена иначе, чем кажется.

Автор залил ролики, набрал 16 000 просмотров — и получил 47 центов встроенной монетизации. Реальные деньги были в другом.

Есть нюансы с верификацией, голосом в роликах и законодательством РФ, которые ломают большинство с…

➡️ Читайте на сайте: https://aff.top/blog/gaid-kak-zarabotat-pervye-dengi-na-pornhub

🧠 Ещё больше инсайтов → в канале AFF.top

Your creative's weight is a bidding variable

We think of creative as a quality lever, not an auction one. But creative latency — how long your ad takes to load and render — feeds back into whether you win and whether the win counts.

1. Many publishers measure render time and viewability of served ads, and persistently slow creatives can be deprioritized or filtered before the auction even reaches you.
2. A heavy creative that wins the auction but renders after the user scrolls past produces a non-viewable, sometimes non-billable impression — you paid the auction cost for nothing.
3. The render rate (the third stage of win-rate decomposition) is partly a creative-weight problem, not just a technical one.

So the same audience and bid can yield very different effective CPMs depending on whether your creative is 40KB or 400KB, because the heavy one loses the silent latency competition that runs alongside the price one.

Why it matters: a lighter creative can raise effective win rate and lower effective CPM without touching a bid, by clearing latency filters and rendering before the impression goes stale. When optimizing programmatic performance, weigh the creative payload as seriously as the bid price — in fast-timeout, viewability-gated environments, the two are part of the same optimization.

Second-price residue: why some of your auctions still aren't first-price

The market 'moved to first-price' around 2017–2019, but that was never universal, and the residue matters because your shading logic assumes an auction type that may be wrong.

1. Plenty of inventory still clears in second-price or hybrid auctions — certain private deals, some ad servers, and exchanges that kept second-price for specific paths.
2. If your DSP shades a bid assuming first-price (lowering it toward expected clearing), but the auction is actually second-price, you've shaded a bid that would have been protected by the second-price mechanism anyway — needlessly lowering your win probability for no price benefit.
3. Conversely, treating a first-price auction as second-price means bidding true value and systematically overpaying.

The DSP has to detect or be told the auction type per path. Exchanges signal it in the bid request (an auction-type field), but the signal isn't always present or honest, and 'audit your auction discrepancies' became a real discipline precisely because the declared type and the observed clearing behavior didn't always match.

Why it matters: shading is only correct when the auction type is known. Mismatches cost you both ways — lost winnable impressions or silent overpayment. Audit the auction-type field against observed clearing prices per path; where they disagree, your shading is solving the wrong problem and your CPMs are paying for it.

First-price auctions removed the anchor — and buyers never replaced it

When the open exchange moved to first-price (you pay exactly what you bid, versus second-price where you paid one cent above the runner-up), the industry framed it as a transparency win. The deeper change was the loss of a free reference price.

The mechanism:
— Under second-price, the auction itself told you the market's valuation: the price you paid was the next-highest bid. That was a continuous, free signal of competitive demand.
— Under first-price, the clearing price equals your own bid. The auction reveals nothing about what others would have paid.
— So every buyer must now estimate the clearing distribution privately, from their own win/loss history — data that is censored, because you never see the prices of auctions you lost.

This is why two DSPs can pay wildly different CPMs for identical inventory: each is reconstructing a hidden price surface from a different, biased sample of its own outcomes.

Implication: bid shading exists precisely to rebuild the anchor that second-price gave away for free. The quality of your shading is the quality of your private price model — and a thin spender on a segment has a worse model than a whale, structurally.

Why it matters: in first-price, information asymmetry is the durable edge. Buyers with more volume on a path estimate its clearing price better and overpay less — a compounding advantage that has nothing to do with creative or targeting.