86.5Leader
Frontier model index · October 2026
Frontier AI Models
- Frontier size
- 10 models
- Supported
- 10 models
- Estimated
- 0 models
- Leader to cutoff
- 11.2 pts
Kimi K3 is #18 overall, 5.4 points below the top-10 cutoff. Its position is Supported. The frontier is the current top 10 in each ranking view on the public board. Supported means we have enough independent evidence to stand behind the row. Estimated means the rank is visible and the uncertainty is wider.
Live index
Which models are on the frontier now?
The table orders public point estimates for the selected capability. Status badges describe the evidence; conditional score ranges do not establish rank confidence. Readers choosing between two labs' top models usually start from GPT-6 Astra compared with Claude Opus 5.5, which reads each category separately.
Cutoff: 75.3
| Rank | Model | Status | Score | Gap | Output / 1M | 90% range |
|---|---|---|---|---|---|---|
| 01 | Claude Opus 5.5Anthropic · Proprietary | Supported | 86.5 | Leader | $20 | 81.7–91.3 |
| 02 | GPT-6.1 SolOpenAI · Proprietary | Supported | 84 | -2.5 | $10 | 80.1–88.0 |
| 03 | GPT-6 AstraOpenAI · Proprietary | Supported | 83.9 | -2.6 | $50 | 79.4–88.3 |
| 04 | Claude Fable 5.1Anthropic · Proprietary | Supported | 83.2 | -3.3 | $50 | 79.9–86.5 |
| 05 | Claude Sonnet 5.5Anthropic · Proprietary | Supported | 82.1 | -4.4 | $10 | 76.5–87.8 |
| 06 | Claude Fable 5Anthropic · Proprietary | Supported | 81.3 | -5.2 | $50 | 78.4–84.2 |
| 07 | Gemini 4 ArgonGoogle · Proprietary | Supported | 81 | -5.5 | $10 | 78.1–83.8 |
| 08 | Claude Opus 5Anthropic · Proprietary | Supported | 79.6 | -6.9 | $25 | 76.8–82.4 |
| 09 | GPT-6 SolOpenAI · Proprietary | Supported | 78.4 | -8 | $10 | 75.3–81.7 |
| 10 | GPT-5.6 SolOpenAI · Proprietary | Supported | 75.3 | -11.1 | $20 | 72.6–78.1 |
#ModelScore
GPT-6.1 SolOpenAI · range 80.1–88.0Supported
84-2.5 pts
GPT-6 AstraOpenAI · range 79.4–88.3Supported
83.9-2.6 pts
83.2-3.3 pts
82.1-4.4 pts
81.3-5.2 pts
81-5.5 pts
79.6-6.9 pts
GPT-6 SolOpenAI · range 75.3–81.7Supported
78.4-8 pts
GPT-5.6 SolOpenAI · range 72.6–78.1Supported
75.3-11.1 pts
Method
Where we draw the line
Frontier membership means the ten highest public point estimates in the selected view. Supported and Estimated describe the evidence behind each row. The ranges are conditional on scoring assumptions; future-score and rank coverage have not been validated.
Multimodal has no public-board ranking, so it has no view here. Missing rows are not filled in.
- 01
Rank
Order ranked models by the selected public score.
- 02
Cut
Keep the first ten rows.
- 03
Label
Mark each row Supported or Estimated.
Cost check
The price of frontier performance
The top score and the cheapest acceptable score are often different buying decisions. This view keeps the evidence status attached to both. What the top of the table costs is clearest in one pair: Claude Fable 5.1 versus GPT-6 Astra puts both models' scores beside their API prices.
Upper-left is better
Scores use the selected frontier view. Prices are official output-token rates in USD per million tokens.
Lowest priced frontier API
GPT-6.1 Sol
$10 / 1M out
Baseline ledger
What changed this month
No model released this month currently sits inside the overall top 10.
Questions
Frontier model questions, answered from the data
Which are the frontier models right now?
As of October 2026, the frontier is the overall top 10 on this page: Claude Opus 5.5, GPT-6.1 Sol, and GPT-6 Astra lead the ranking, and 10 of the 10 hold Supported positions. The full list above updates as scores and evidence change.
What is the difference between frontier models and foundation models?
Foundation model describes a class: large models trained broadly so they can be adapted to many tasks. Frontier model describes a rank: the small, shifting subset of foundation models at the top of current measured capability. Every frontier model is a foundation model; most foundation models are not frontier. Likewise, LLM names the architecture family — frontier names where a model currently stands.
Why are they called frontier models?
The frontier is the outer edge of demonstrated AI capability — a moving boundary, not a fixed specification. A model is on the frontier only relative to everything else available at the time, which is why this index is dated, re-scored as benchmark evidence lands, and records when models enter or leave.
How big are frontier models?
Mostly undisclosed. Closed labs stopped publishing parameter counts, and recent open-weight frontier entries that do publish them are mixture-of-experts designs with totals in the hundreds of billions of parameters. Size is a weak proxy for capability, so this index tracks what is measurable instead: benchmark scores with exact sources, and what the model costs to run.