This is the Saturday mini-roundup: three stories that shaped the week for anyone using AI tools. DeepSeek shipped its flagship V4-Pro model and announced its first big price hike, with a new peak/off-peak billing system. GLM-5.3 arrived from Z.ai as an open-weights coding model with standout cyber-security scores, though the weights are deliberately delayed. And ChatGPT ads expanded to more markets as Google’s Gemini app passed a billion users. We’ve sorted the week’s news for beginners. This roundup is based on official announcements, vendor documentation, and news reports — we did not test these tools hands-on.
DeepSeek V4-Pro Is Here — Along with Its First Big Price Hike
DeepSeek shipped its flagship V4-Pro model (DeepSeek-V4-Pro-0813) to its API on Aug 13, 2026, alongside its first major price hike and a new peak/off-peak billing system Pandaily. For beginners, this matters because DeepSeek has been the default cheap AI for API users — and this is its first big hike, landing just as OpenAI cut prices to pressure Chinese challengers Caixin Global. We covered OpenAI’s side of the market split in our briefing on OpenAI’s $7B share sale.
When Is DeepSeek Cheapest to Use?
Off-peak hours cost exactly half of peak pricing, and the new billing takes effect 16:00 UTC Sat Aug 16, 2026 DeepSeek. Peak windows are 01:00–04:00 and 06:00–10:00 UTC; everything else is off-peak at half price DeepSeek. If you run batch jobs or any non-urgent API work, schedule it outside those two peak windows and you pay half. This is the first big test of dynamic pricing in AI APIs, so watch how it plays out.
DeepSeek V4-Pro vs GPT-5.6 Luna vs Kimi K3 — Picking a Budget Model
The budget sibling V4-Flash is still the cheapest DeepSeek option at peak prices of $0.014 cached input, $0.44 uncached input, and $1.32 output per 1M tokens DeepSeek. GPT-5.6 Luna dropped to $0.20 input and $1.20 output per 1M after an ~80% cut Caixin Global. Kimi K3 is another strong open-weights option — compare all three side by side in our Comparison Database.
Verdict: Should Beginners Panic About DeepSeek’s Price Hike?
No — but costs are changing, so adjust your habits rather than panic. Shift batch work to off-peak hours and compare prices before switching providers, because the gap between DeepSeek and US models has narrowed significantly Caixin Global. V4-Pro’s new peak rates are $0.044 cached input, $1.32 uncached input, and $3.96 output per 1M tokens, versus old flat rates of $0.003625, $0.435, and $0.87 — roughly a 12x jump on cache-hit input and about 4.5x on output Caixin Global Pandaily.
FAQ
Short answers to the questions beginners ask most about DeepSeek’s new pricing.
When do DeepSeek’s off-peak hours start?
Off-peak billing takes effect 16:00 UTC Sat Aug 16, 2026, which is midnight Beijing time Aug 17 DeepSeek. Peak windows are 01:00–04:00 and 06:00–10:00 UTC; everything else is off-peak at exactly half the peak price DeepSeek. If your usage is flexible, schedule heavy jobs outside those windows.
Does the DeepSeek price hike affect ChatGPT or Claude?
No — this is DeepSeek’s own API pricing, and it doesn’t change what ChatGPT or Claude charge Caixin Global. In fact, US models moved the other direction this summer: OpenAI cut GPT-5.6 Luna about 80% to $0.20 input and $1.20 output per 1M tokens Caixin Global. The two markets are diverging.
Is DeepSeek still the cheapest AI API for beginners?
Still competitive, especially if you use off-peak hours or the V4-Flash sibling, but the gap has narrowed Pandaily. V4-Flash peak prices are $0.014 cached input, $0.44 uncached input, and $1.32 output per 1M tokens DeepSeek. Compare before committing to a provider.
GLM-5.3: The Open-Weight Coding Model with Cyber Skills
Z.ai launched GLM-5.3 on Aug 14, 2026, a coding upgrade built through post-training on the same base model as GLM-5.2 Z.ai. For beginners, the headline is that the model scores extremely well on cyber-security benchmarks, and the open weights are deliberately delayed about two weeks Z.ai. It connects to our earlier guide on running AI agents on your laptop.
What Does “Open Weights” Mean?
Open weights means the model’s parameters are downloadable, so you can study them, fine-tune them, and run the model locally instead of only through an API Z.ai. This matters for cost and privacy: you pay for compute, not per-token API fees, and your data stays on your machine. GLM-5.3’s weights are promised roughly two weeks after launch, around Aug 28, once safety evaluation and hardening are complete Z.ai.
GLM-5.3 vs Qwen3.8-27B vs Claude — Which Should Beginners Use?
Qwen3.8-27B is the open option with weights you can download and run today, while Claude is the closed frontier reference for comparison Hugging Face. GLM-5.3’s coding scores are strong: roughly +50% over GLM-5.2 on Z.ai’s in-house Code Bench, and open-source SOTA on Terminal Bench 3.0 at 28.3 versus 4.6 Z.ai. Use our Comparison Database to see how they stack up for your use case.
Verdict: Should Beginners Wait for GLM-5.3’s Open Weights?
Wait if you want free weights, which are promised around Aug 28 after safety review Z.ai. Pay from $12.6 per month if you want GLM-5.3 in a coding agent today Z.ai. Either way, GLM-5.3 is beginner-relevant only if you code — if you don’t, the other two stories in this roundup matter more.
FAQ
Short answers to the questions beginners ask most about GLM-5.3’s launch.
Is GLM-5.3 free to use?
The weights are not out yet — they’re promised about two weeks after launch, around Aug 28, after safety evaluation Z.ai. Paid access is available today starting with the GLM Coding Plan at $12.6 per month for Lite Z.ai. If you want free weights, you wait; if you want it now, you pay.
Why is Z.ai delaying the open weights?
Z.ai says the delay is for safety evaluation and hardening before release Z.ai. The model tops cyber benchmarks — CyberGym 84.5% ahead of GPT-5.6 Sol at 83.6% — so gating the release is the point Z.ai. This is a deliberate safety decision, not a technical problem.
How can I try GLM-5.3 today?
Sign up for the GLM Coding Plan — Lite at $12.6 per month, Pro at $56 per month, or Max at $117.6 per month — and use it with the ZCode agent Z.ai. ZCode routes inside Claude Code, Cline, or Kilo Code Z.ai. You get the model in your existing coding workflow.
ChatGPT Ads Expand — and Gemini Just Passed 1 Billion Users
OpenAI expanded ChatGPT Ads to the UK, Mexico, Brazil, Japan, and South Korea, and on the same day Google announced the Gemini app passed 1 billion monthly users OpenAI TechCrunch. Ads appear only for logged-in adults on Free and Go tiers, while Plus, Pro, Business, and Edu stay ad-free OpenAI. The same week ChatGPT gained restaurant booking, as we covered in our briefing on ChatGPT restaurant booking.
How Do I Turn Off Ads in ChatGPT?
Free users can opt out of ads entirely, but in exchange you get fewer daily free messages OpenAI. If you don’t want to opt out, you have controls: dismiss ads, view your ad history, delete ad data with one tap, and a personalization toggle OpenAI. The ads are labeled “sponsored” and separated from answers OpenAI.
ChatGPT Free with Ads vs Gemini vs Perplexity — Where Should You Chat?
Claude markets itself as ad-free, Gemini has 1 billion users, and Perplexity already runs ads — so the free-tier experience varies widely TechCrunch. ChatGPT’s ads are currently limited to Free and Go tiers in nine markets, with paid tiers ad-free OpenAI. Check our Comparison Database to see the trade-offs side by side.
Verdict: Should You Pay to Escape ChatGPT Ads?
Free users in the five new markets should keep working for free and use the opt-out plus the dismiss and personalization controls OpenAI. Don’t pay just to dodge ads — try Gemini free first, since it just crossed 1 billion monthly users TechCrunch. Revisit the decision if ads feel intrusive, but a paid ChatGPT plan isn’t the only escape hatch.
FAQ
Short answers to the questions beginners ask most about ChatGPT ads and Gemini’s milestone.
Will I see ads in my ChatGPT chats?
Only if you’re a logged-in adult user on the Free or Go tiers, and only in the US, Canada, Australia, NZ, UK, Mexico, Brazil, Japan, and South Korea OpenAI. Plus, Pro, Business, and Edu accounts never see ads OpenAI. The pilot began in the US on Feb 9, 2026, then expanded to Canada, Australia, and New Zealand ghacks.
Can I turn ads off in ChatGPT completely?
Yes — you can opt out entirely in exchange for fewer daily free messages OpenAI. Or you can dismiss individual ads, view your ad history, and toggle off personalization without losing messages ghacks. The controls are designed so you have options short of paying.
Does ChatGPT share my chats with advertisers?
No — advertisers get aggregate stats only, like views and clicks, never your chats, history, or memory OpenAI. Past chats can influence targeting only if you enable ad personalization OpenAI. Your conversations stay private either way.
What to watch next week (Aug 16–22)
Next week is when several of this week’s announcements go live. A pricing experiment starts, an open-weights rollout lands, and NVIDIA’s earnings set the tone for AI costs across the industry. Here’s what to track.
- DeepSeek peak/off-peak pricing goes live Sat Aug 16 at 16:00 UTC — the first big test of dynamic AI pricing. DeepSeek
- GLM-5.3 open weights roll out around Aug 28, pending safety review — paid coding-plan access is available now. Z.ai Z.ai
- Qwen3.8-27B hosted API — the open weights are out now on Hugging Face; Qwen Cloud’s hosted access has not launched yet. Hugging Face
- NVIDIA Q2 FY27 earnings Wed Aug 26 — guidance sets the tone for AI stocks and cloud pricing. Finance Calendar
- Debian’s community vote on LLM-written code opens Aug 15 and closes Aug 28. Debian
This story was produced by our automated pipeline — track what’s coming next at our cron pipeline.
Back to all posts
