claude is back!!
Updated
So after claude opus 5.5 was released yesterday, its api pricing dropped from $5/$25 to $4/$20. I also saw a lot of opus 5.5 demos showing its capabilities from most of my oomfs. The vibe was more like opus 4.5 back then: strong positive sentiment across the timeline, with people satisfied with its capabilities. But the important factors when choosing a model for me as a brokie here are pricing and usage. As you know, fable is really strong, if not the most intelligent model, head-to-head with the astra model by OpenAI. My only concern is that both models, despite currently being the most intelligent models here, drain users’ weekly usage very fast. Thus, I don’t find these types of models efficient as daily drivers (due to being broke too). I saw some people say that claude’s $20 subs plan is usable enough to drive opus 5.5. I re-subbed to claude again after abandoning my $100 claude max plan a few months ago due to codex being good and generous enough for its subs plan.
So I did some vibecalculation with the help of deepseek v4.1 flash (too oddly specific lol) and ccusage combined. Here is the result so far based on my remaining quota and tokens spent…
usage since yesterday
| Source | Date | Model | Input | Output | Cache Write | Cache Read | Value (USD) |
|---|---|---|---|---|---|---|---|
| Claude Code | 2026-09-23 | claude-opus-5-5 | 454 | 276,216 | 605,638 | 85,684,183 | $27.5081 |
| Claude Code | 2026-09-23 | claude-sonnet-5 | 8 | 1,689 | 10,584 | 26,755 | $0.0487 |
| Claude Code | 2026-09-24 | claude-opus-5-5 | 2,142 | 870,051 | 5,543,593 | 248,127,298 | $109.4553 |
| Claude Code subtotal | All | Claude | 2,604 | 1,147,956 | 6,159,815 | 333,838,236 | $137.0121 |
| omp | 2026-09-24 | claude-opus-5-5 | 334 | 104,233 | 601,255 | 30,895,618 | $13.0752 |
| Claude-only total | All | Claude | 2,938 | 1,252,189 | 6,761,070 | 364,733,854 | $150.0873 |
live quota status (as i write this note)
| Metric | Value |
|---|---|
| Checked at | 2026-09-24 23:33 UTC |
| 5-hour usage | 100%; resets 2026-09-25 00:20 UTC |
| 7-day usage | 53%; 47% remaining |
| 7-day window | 2026-09-18 19:00 → 2026-09-25 19:00 UTC |
| Quota breakdown | Claude Code 99%; Chats 1% |
claude-only api-equivalent value estimate on claude pro subs
| Method | Weekly | Monthly | vs $20/mo |
|---|---|---|---|
| Full quota: $150.0873 / 0.53 | $283.18 | $1,226.18 | 61.31x |
| Observed burst: $150.0873 / 1.643 days x 7 | $639.38 | $2,768.51 | 138.43x |
monthly = weekly x 4.33. Quota estimates assume the same usage mix. burst elapsed time starts at 2026-09-23 08:07:20 UTC. the 5-hour cap makes the burst rate non-sustainable. values are API-equivalent estimates, not subscription charges.
So far, since yesterday:
443.3M processed tokens
1,960 model calls
1,037 tool calls
5h window: 100%
Weekly window: 53%It also appears to fix claude opus 5’s le claudious sloppish jargon wording, which is unreadable for a mere human peasant like me. The wording from opus 5.5 felt more natural to me. Also remember, based on the data I gave above, how usable and cheap claude opus 5.5 is compared to previous ant models and how good it is (way better than fable 5.1, a significant improvement over opus 5).
For visual design, I also believe ant improved its model’s taste a lot. A few examples here:
- https://x.com/kevin_t_ngo/status/2102437977435893771?s=20
- https://x.com/other__reality/status/2102514581684052169?s=20
- https://x.com/donaldjewkes/status/2102801274173587569?s=20
- https://x.com/devteamdrew/status/2102436464323661880?s=20
- https://x.com/pleometric/status/2103082510607610023?s=20
What I want to emphasize here is usability and the usage allowance alone on claude subs. Previously, claude fable was very good but not really that usable given how expensive it was. Now opus 5.5 is literally priced ~20% lower than the usual opus api rates while being better than fable. The usage allowance on claude pro subs is also generous enough, which is very satisfying!
anthropic might achieved efficiency
anthropic is back!
unless they nerf and quantize their model into haiku 5 next week hehehe