Home / Compare / Best AI models
The best AI models right now, and what they cost
Three models share the top of Epoch AI’s independent Capabilities Index: Claude Opus 5.5, GPT-6 Astra and GPT-6.1 Sol. Their scores sit within each other’s margins of error, but their prices don’t: per million tokens, GPT-6.1 Sol costs $2.00 / $10.00, Opus 5.5 $4.00 / $20.00 and Astra $10.00 / $50.00. Here are the top 12, the best value at each price, the best open models and what developers actually use.
Rankings from Epoch AI and OpenRouter, prices from each company’s pricing page, checked October 18, 2026
The best AI for each job
Also: studying, designers, business, and Which AI should I use? to pick by budget.
The top 12 AI models
The Epoch Capabilities Index combines scores from many benchmarks into one scale; a model scores higher when it does better on harder tests. Each score has a confidence range, and models whose ranges overlap are effectively tied.
| # | Model | Maker | Score | Range | Released | Input / output |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | Anthropic | 166.8 | 163.5–171 | Sep 22, 2026 | $4.00 / $20.00 |
| 2 | GPT-6 Astra | OpenAI | 166.4 | 163.3–171.2 | Sep 3, 2026 | $10.00 / $50.00 |
| 3 | GPT-6.1 Sol | OpenAI | 166.3 | 163.2–170.8 | Sep 29, 2026 | $2.00 / $10.00 |
| 4 | Claude Fable 5.1 | Anthropic | 164.7 | 161.9–168.5 | Sep 1, 2026 | $10.00 / $50.00 |
| 5 | Gemini 4 Argon | 164.6 | 161.7–168.7 | Sep 30, 2026 | $2.00 / $10.00 | |
| 6 | Claude Sonnet 5.5 | Anthropic | 164.4 | 161.3–168.3 | Sep 28, 2026 | $2.00 / $10.00 |
| 7 | Claude Opus 5 | Anthropic | 162.6 | 160–165.8 | Jul 24, 2026 | $5.00 / $25.00 |
| 8 | GPT-6 Sol | OpenAI | 162.6 | 160–165.8 | Sep 22, 2026 | $2.00 / $10.00 |
| 9 | Claude Fable 5 | Anthropic | 162.1 | 159.5–166.3 | Jun 9, 2026 | — |
| 10 | GPT-5.5 Pro | OpenAI | 162 | 159–165.1 | Apr 23, 2026 | — |
| 11 | GPT-5.6 Sol | OpenAI | 161.7 | 159.5–165 | Jul 9, 2026 | $4.00 / $20.00 |
| 12 | GPT-5.6 Terra | OpenAI | 159.4 | 156.8–162.2 | Jul 9, 2026 | $2.00 / $12.00 |
Source: Epoch AI, “Capabilities & benchmarking”, epoch.ai/benchmarks, licensed CC BY 4.0; scores as of October 18, 2026. Prices are the makers’ API list prices; Gemini 4 Argon’s introductory price doubles on December 1, 2026.
Best value: top results for less
At the top. GPT-6.1 Sol ties for first at $2.00 / $10.00 per million tokens, a fifth of GPT-6 Astra’s price. Claude Sonnet 5.5 and Gemini 4 Argon cost the same as Sol and score within a couple of points of it. Claude Opus 5.5 costs twice as much as Sol and has the highest score, by a margin smaller than its range.
In the cheapest tier. Claude Haiku 5.5 and GPT-6 Luna both cost $0.10 / $0.50. Haiku scores 159 and Luna 156.3, though their ranges overlap.
| Model | Score | Input / output |
|---|---|---|
| Claude Haiku 5.5 | 159 | $0.10 / $0.50 |
| Gemini 3.8 Flash | 157.2 | $0.75 / $3.75 |
| GPT-6 Luna | 156.3 | $0.10 / $0.50 |
| DeepSeek V4.1 Flash | 155.6 | $0.30 / $1.20 |
| Grok 4.7 | 154.4 | $2.00 / $6.00 |
DeepSeek’s price is its peak-hour rate; off-peak costs half. Work out a month at your own usage with the cost calculator.
Capability for the price, every ranked model
Epoch Capabilities Index (up) against the blended price per 1M tokens, 3 input to 1 output (right, log scale). Dots on the line have no model that is both cheaper and higher scoring. Hover or tap a dot for its numbers.
Show the numbers (16 models)
| # | Model | Company | Score | Blended per 1M | Value line |
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | Anthropic | 166.8 | $8.00 | Yes |
| 2 | GPT-6 Astra | OpenAI | 166.4 | $20.00 | |
| 3 | GPT-6.1 Sol | OpenAI | 166.3 | $4.00 | Yes |
| 4 | Claude Fable 5.1 | Anthropic | 164.7 | $20.00 | |
| 5 | Gemini 4 Argon | 164.6 | $4.00 | ||
| 6 | Claude Sonnet 5.5 | Anthropic | 164.4 | $4.00 | |
| 7 | GPT-6 Sol | OpenAI | 162.6 | $4.00 | |
| 8 | Claude Opus 5 | Anthropic | 162.6 | $10.00 | |
| 9 | GPT-5.6 Sol | OpenAI | 161.7 | $8.00 | |
| 10 | GPT-5.6 Terra | OpenAI | 159.4 | $4.50 | |
| 11 | Claude Haiku 5.5 | Anthropic | 159.0 | $0.20 | Yes |
| 12 | Gemini 3.8 Flash | 157.2 | $1.50 | ||
| 13 | GPT-6 Luna | OpenAI | 156.3 | $0.20 | |
| 14 | DeepSeek Flash | DeepSeek | 155.6 | $0.525 | |
| 15 | DeepSeek V4 Pro | DeepSeek | 155.3 | $1.98 | |
| 16 | Grok 4.7 | xAI | 154.4 | $3.00 |
Scores: Epoch AI, Capabilities & benchmarking (Epoch Capabilities Index), CC BY 4.0, checked October 18, 2026. Each score carries a margin of error of about ±3 points, so models a few points apart may be level. Prices: each company’s pricing page, standard tier.
The best open-weight models
Open-weight models can be downloaded and run on your own hardware or through any host. The highest-scoring one, Kimi K3, is for non-commercial use only; DeepSeek’s models carry no use restrictions.
| Model | Maker | Score | License | Maker’s API, input / output |
|---|---|---|---|---|
| Kimi K3 | Moonshot | 157.4 | Open weights, non-commercial use only | — |
| DeepSeek V4.1 Flash | DeepSeek | 155.6 | Open weights, no use restrictions | $0.30 / $1.20 |
| DeepSeek V4 Pro 0813 | DeepSeek | 155.3 | Open weights, no use restrictions | $1.32 / $3.96 |
| DeepSeek V4 Flash 0731 | DeepSeek | 154.5 | Open weights, no use restrictions | — |
| GLM-5.3-Flash | Z.ai | 153.2 | Open weights, no use restrictions | — |
| GLM-5.2 | Z.ai | 151.9 | Open weights, no use restrictions | — |
Source: Epoch AI, “Capabilities & benchmarking”, epoch.ai/benchmarks, licensed CC BY 4.0. More on DeepSeek: DeepSeek API pricing and DeepSeek vs ChatGPT.
What developers actually use
OpenRouter, a service that routes developers’ requests to hundreds of models, publishes how many tokens each model handles. Cheap, fast models dominate: in the week to October 17, 2026, only one of the top-scoring models, Claude Opus 5.5, made its top 10. OpenRouter notes that token counts show how much a model is used, not which model is best.
| # | Model | Maker | Tokens |
|---|---|---|---|
| 1 | DeepSeek V4.1 Flash | DeepSeek | 35.9T |
| 2 | Step 5 Preview | StepFun | 13.9T |
| 3 | GLM 5.3 Flash | Z.ai | 12.6T |
| 4 | MiMo-V2.6-Flash | Xiaomi | 10.3T |
| 5 | Hy4 preview | Tencent | 7.55T |
| 6 | GPT-6 Luna | OpenAI | 6.29T |
| 7 | Claude Opus 5.5 | Anthropic | 6.05T |
| 8 | Nemotron 3 Ultra (free) | NVIDIA | 5.34T |
| 9 | GLM 5.3 | Z.ai | 5.34T |
| 10 | DeepSeek V4 Flash 0731 | DeepSeek | 4.71T |
By maker, DeepSeek had 24.5% of OpenRouter’s requests in the week beginning October 5, 2026, OpenAI 18.6%, Google 17.8% and Anthropic 3.9%. Source: OpenRouter LLM Rankings, licensed CC BY 4.0.
How to read these rankings
- Ties are real. When score ranges overlap, the order can change with the next benchmark run. Pick by price, speed and how the model handles your own examples.
- Benchmarks aren’t your work. An index averages many tests; a model that leads overall can trail on the one task you care about.
- Makers’ own charts favor their models. We rank by Epoch’s independent index and quote makers’ claims only as their claims.
- We don’t test models ourselves or take payment for placement; every number links to its source.
Questions
What is the best AI model right now?
Claude Opus 5.5, GPT-6 Astra and GPT-6.1 Sol are tied at the top of Epoch AI’s Capabilities Index as of October 18, 2026. Of the three, GPT-6.1 Sol costs the least through the API.
What is the best LLM?
A large language model (LLM) is the kind of AI model behind ChatGPT, Claude and Gemini, so the answer is the same: Claude Opus 5.5, GPT-6 Astra and GPT-6.1 Sol share first place on Epoch’s index.
What is the best open-source LLM?
Among open-weight models, Kimi K3 scores highest on Epoch’s index but allows only non-commercial use. The best with no use restrictions is DeepSeek V4.1 Flash, which is also the most used model on OpenRouter.
What is the best AI model for coding?
Anthropic’s Claude Opus 5.5 and OpenAI’s GPT-6.1 Sol are both pitched at coding and sit at the top of Epoch’s overall index; Sol costs half as much per token. Try both on your own code before committing.
Is the most expensive AI model the best?
No. GPT-6 Astra and Claude Fable 5.1 cost $10.00 / $50.00 per million tokens, but GPT-6.1 Sol ties Astra for first at a fifth of the price.
Sources
- Epoch AI, Capabilities & benchmarking and its Capabilities Index (scores, ranges, release dates, licenses), CC BY 4.0
- OpenRouter LLM Rankings (usage by model and maker), CC BY 4.0
- API prices: each maker’s pricing page, via our Claude, OpenAI, Gemini, Grok and DeepSeek pages
- Anthropic: Claude Opus 5.5 and OpenAI’s API changelog (coding claims)