The top of the League table costs 38× the bottom of the top 10.
AI Pro League’s October standings put GPT-6 Astra at the top on capability — at a blended $20.00 per million tokens, roughly 38 times the cheapest model inside the top 10. Here is the whole top 10, priced from our own database.
By AI Pro League · Published
The top AI models in our League standings are separated by 18 points of League Score from first to tenth — and by a factor of 38 in price. That second gap is the one budgets feel. This analysis uses only the numbers in our own model database: League Scores from the published benchmark record, and first-party API prices checked against each maker’s pricing page (the check date sits on every model page). Blended cost is our standard 3:1 input-to-output mix — the method is on the pricing index.
The top AI models, priced
| # | Model | Maker | League Score | $/1M in | $/1M out | Blended |
|---|---|---|---|---|---|---|
| 1 | GPT-6 Astra | OpenAI | 100 | $10.00 | $50.00 | $20.00 |
| 2 | Claude Opus 5.5 | Anthropic | 99 | $4.00 | $20.00 | $8.00 |
| 3 | Claude Fable 5.1 | Anthropic | 97 | $10.00 | $50.00 | $20.00 |
| 4 | GPT-6.1 Sol | OpenAI | 97 | $2.00 | $10.00 | $4.00 |
| 5 | Claude Opus 5 | Anthropic | 95 | $5.00 | $25.00 | $10.00 |
| 6 | Claude Sonnet 5.5 | Anthropic | 88 | $2.00 | $10.00 | $4.00 |
| 7 | GPT-5.6 Sol | OpenAI | 87 | $4.00 | $20.00 | $8.00 |
| 8 | Gemini 4 Argon | Google DeepMind | 86 | $2.00 | $10.00 | $4.00 |
| 9 | GPT-6.1 Terra | OpenAI | 83 | $1.00 | $5.00 | $2.00 |
| 10 | DeepSeek V4.1 Flash | DeepSeek | 82 | $0.30 | $1.20 | $0.52 |
The median blended price inside the top 10 is $8.00 per million tokens, against $3.00 across all 32 scored models with first-party API pricing. The top of the table really is a different price tier — with one exception, below.
What 100 million tokens would cost
Blended rates become real money at volume. At the same 3:1 mix, 100 million tokens through three points of the top 10 costs:
| Model | Cost per 100M tokens |
|---|---|
| GPT-6 Astra | $2,000.00 |
| Claude Fable 5.1 | $2,000.00 |
| DeepSeek V4.1 Flash | $52.50 |
Run your own mix — coding traffic is rarely 3:1 — through the API cost calculator before treating any of these as your number.
The value corner of the table
DeepSeek V4.1 Flash is the exception that proves the pricing rule: it sits at #10 overall with a League Score of 82 and costs $0.52 blended — and it is the top 10’s only open-weights entry.
Ranked purely on score per dollar, our value ranking is led by Claude Haiku 5.5 (League Score 66 at $0.20 blended); GPT-6 Luna (League Score 55 at $0.20 blended); DeepSeek V4.1 Flash (League Score 82 at $0.52 blended). The full budget shortlist lives on the best budget AI models page, and the monthly price floor for every maker is tracked in the cheapest capable AI models.
Read the gap honestly
Two caveats belong on the record. First, coverage: a League Score blends up to five benchmark components, and the top 10’s scores rest on between 40% and 90% of that record — each model page shows exactly which components its score uses, and the methodology explains the blend. Second, the head-to-head view: Fable 5.1 vs GPT-6 Astra shows the top two at the same price are not the same model. Price gaps are real; what they buy still has to be read model by model.