# Claude or Codex for $200

Codex’s API-equivalent value has dropped, but cheaper models like Sol 6.1 may still deliver more work for the budget. Meanwhile, Claude has become a stronger all-in-one subscription again.

Author: Maxim Tolmachev
Published: 2026-10-08
Language: en
Canonical: https://maximtolmachev.com/blog/claude-or-codex-for-200/
Original source: https://t.me/mtolm/105

I watched Theo’s latest breakdown of Codex’s value, and the numbers are pretty unpleasant.

By his estimate, until recently the $200 subscription provided around $12,000 a month in value when usage was recalculated at API rates. With all limits fully used and frequent resets, it reached $24,000. After the usage accounting changed, he got around $570 for a week, or roughly $2,000–2,500 a month.

Even on the $500 plan, using Astra Ultrafast, Theo gets around $4,000 in API-equivalent value per month. You can burn through the weekly limit in literally two hours of generation 🥴

But I wouldn’t choose a subscription based on that number alone. The API equivalent also depends on the model’s price: the more expensive its tokens are, the more “dollars” the counter will show for the same work.

In Theo’s own Terminal-Bench test, GPT-6.1 Sol xhigh practically matched Opus 5.5 in performance. The cost per task was $0.96 versus $13.11—roughly 14 times less.

So a drop in the dollar equivalent doesn’t tell you how many fewer tasks you’ll be able to complete. With Sol 6.1, a smaller budget may cover more work. This is the model that has become my main one now.

At the same time, Claude has become more interesting as a subscription again. Opus 5.5 for complex tasks, the new Haiku 5.5 for everyday work. You can build a solid setup within one subscription and spend less of the expensive model on routine tasks.

Anthropic is also adding monthly API credits for the $200 Max 20x plan starting October 7. You need to activate them by linking Console. This is a separate budget for apps and agents; it doesn’t increase the interactive Claude Code limits.

I’m also keeping the restrictions in mind. On Max, Fable can consume up to 50% of the total weekly limit. And Claude subscription limits shouldn’t be treated as a universal API for Hermes, OpenClaw, or a service for other users. External tools have separate access and billing rules, while your own user-facing service requires the API.

If you’re choosing one subscription for working in Claude Code, I’d seriously consider Claude right now. If Codex with Sol 6.1 is enough for you, I don’t see any reason to leave it urgently.

In the next post, we’ll go through a detailed atlas of models and effort levels for different tasks.

Are your current subscription limits enough for you?
