A July Full of Resets: A Light AI User Reviews the Subsidy War

Every time a reset landed in July I used it to the last drop, and the ledger says 57.5x. Plus notes on the subsidy war, Deepseek's price hike, and my bet on the next reset.

On this page

What I’m already actively doing
That Deepseek chart may be obsolete soon
On resets themselves

A light AI user here, reviewing how this “July full of resets” went.

I pay 100 a month for Claude and 100 for Codex, and every time a reset landed I used it to the last drop. As the screenshot shows, that came out to 57.5x of value (the price calculation already excludes the local qwen models in there).

Screenshot of the ccusage table for all of July, totaling 233 million input tokens and 61.19 million output tokens, worth $11,501.51; the Claude row accounts for $7,415.12 and the Codex row for $4,086.39

I’m honestly looking forward to either OpenAI or Anthropic going public, because that’s when the market gets a chance to see what tokens actually cost.

No idea how long the subsidy war will run. GPT announced another price cut yesterday, and Kimi and Deepseek keep pushing their benchmarks closer while driving prices down. I can’t read the situation clearly, and even if I could, ordinary folks like us can’t change anything about it.

What I’m already actively doing

Keeping the harness on a diet (cut 40% of the always-injected tokens this week), routing some very simple one-shot tasks to local models overnight, or sending them through the all-you-can-eat Gemini subscription (antigravity CLI).

Further routing I plan to do: non-sensitive information, or via local…

That Deepseek chart may be obsolete soon

That widely circulated “Deepseek price-performance kill line” chart may be obsolete very soon. Deepseek announced it’s about to raise API prices substantially.

Screenshot of a post from the account Jukan on X saying DeepSeek plans to raise overall API pricing in the near future with a significant increase expected, below it the announcement banner from the DeepSeek platform usage page, along with fields showing a $19.75 balance and $0.24 total cost

I can picture Luna and Terra grinning.

On resets themselves

From my (unprofessional) research, OpenAI’s resets aren’t as generous as they look: the seven-day window restarts after the reset, while Anthropic keeps the seven-day window fixed. The latter effectively gives you 100% of your quota over a shorter window after the reset. And looking at when past resets were announced, they always land right as the previously issued banked resets are about to expire.

So I’m putting this on record: there will definitely be a reset on 8/12-14, and you can all come back and check. Why? Because Tibo is sly, and always times the reset to hit right on the banked reset expiry date. The last banked reset expires 8/13.

Another gut feeling: the weekly quota on Codex’s 100 USD plan seems smaller than Claude’s 100 USD one. I haven’t tallied token usage, so this is purely a feeling. Anyone else feel the same? Not sure whether it’s because Opus 5 is that durable, or because I run so many browser and computer-control tasks on Codex.

One more thing worth noting: Anthropic’s latest post lists only Fable, Opus, and Sonnet, with no mention of Haiku at all. Sonnet looks like it’s taking over the slot Haiku used to hold, which is a price increase nobody’s calling a price increase. Haiku will probably stop at 4.5.

Anthropic, so stingy.

When the quota runs out, go leave a comment on X praying to Tibo. When the reset comes, /model sol max. When it’s truly gone, pull out the credit card and cry while paying the 200.