Opus 5 Against the Field: When You Do Not Need the Best Model in the World
The top model costs $10 per million tokens blended. Kimi K3 delivers 93 percent of its intelligence for 60 percent of the price, and GLM-5.2 costs eleven times less. Run this calculation before you move production over.
Opus 5 is now the highest-scoring model in the independent Intelligence Index. The question I ask at every launch is not whether it is the best. It is what I actually need the best model for.
What intelligence costs right now
Prices below are the blended 3:1 input-to-output figure from Artificial Analysis, which sits close to what typical usage bills at:
| Model | Intelligence | Price / 1M |
|---|---|---|
| Claude Opus 5 (max) | 61 | $10 |
| Claude Fable 5 | 60 | $20 |
| GPT-5.6 Sol | 59 | $11.25 |
| Kimi K3 | 57 | $6 |
| Claude Opus 4.8 | 56 | $10 |
| GLM-5.2 (max), open weights | 51 | $0.90 |
Three things fall out of that table. Opus 5 halved the price of the frontier: same rank as Fable 5, twice as cheap. Kimi K3 gives you 93 percent of the leader’s score at 60 percent of the cost. And GLM-5.2, with open weights, costs eleven times less while still scoring 51.
Where those ten points actually live
The gap between 61 and 51 is not spread evenly across your workload. You will not see it summarising an email, rewriting a paragraph or pulling fields out of an invoice. You will see it in multi-step tasks, in code you did not write, and anywhere the model has to notice that it just made a mistake.
How to run the numbers yourself
Take a hundred real prompts from your last week, run them through both models, and count how many answers needed a fix. If the cheaper model needs fixing five percent of the time and costs four times less, the decision makes itself. If it needs fixing thirty percent of the time, you are paying for the expensive model anyway, just in your own hours.
Effort control adds a third option that did not exist before: instead of switching models, drop the dial on the one you already use. That decision now sits with you, on every single request.
