AI · LEADERBOARD

The AI Leaderboard

Q: What is the best AI model right now?

As of 5 September 2026, Claude Fable 5.1 (max with fallback) by Anthropic is the highest-ranked model on our daily intelligence index. The leaderboard re-ranks every day, so check back for the current pick.

Q: What is the best open-source AI model?

Kimi K3 (max) is currently the strongest open-source model in our ranking, the top pick for teams that want to self-host or keep data sovereign.

Q: Which AI model has the best price-to-performance?

GPT-5.6 Luna (max) currently scores highest on our value metric, which weighs intelligence and speed against the cost per million tokens.

Q: Which AI model has the largest context window?

Kimi K3 (max) leads on context window at 1M tokens, useful for long documents, large codebases and extended conversations.

Q: How often is the leaderboard updated?

Every day at 04:00 Europe/Amsterdam. We pull fresh public benchmark data and re-rank, with no smoothing and no manual edits.

Q: How are the AI models ranked?

We aggregate public benchmarks into an intelligence index, then score value as intelligence weighed against output speed and cost per million tokens. Output cost is weighted more heavily because real usage is output-heavy.

Q: Is the SevenLab AI leaderboard free and independent?

Yes. The ranking is free, published under Creative Commons, and fully independent. We take no vendor payments to place models, these are the picks our own teams ship with.

Daily ranking of the strongest AI models. No vendor spin, no marketing pages, just the picks teams actually ship with.

As of 5 September 2026, the top-ranked AI model is Claude Fable 5.1 (max with fallback) by Anthropic. The strongest open-source model is Kimi K3 (max), and GPT-5.6 Luna (max) offers the best price-to-performance. Rankings refresh every day at 04:00 Europe/Amsterdam.

Best closed model

Claude Fable 5.1 (max with fallback)

Anthropic

Intel: 56.8
Speed: 71t/s
Cost in/out: €8.6 / €43.02

State-of-the-art reasoning and complex coding tasks requiring maximum intelligence

Best open model

Kimi K3 (max)

Kimi

Intel: 50.2
Speed: 38t/s
Cost in/out: Provider€2.58 / €12.91Sovereign~€132 / hr*

Sovereign on-premise deployments requiring massive context and high intelligence

Best value103.9

GPT-5.6 Luna (max)

OpenAI

Intel: 43.4
Speed: 110t/s
Cost in/out: €0.17 / €1.03

Ultra-low-cost value-tier tasks with high speed and large context

Top 10, side by side

#↑	Model	Vendor	Intelligence	Speed (tok/s)	Cost (€/Mtok)	Context	Best for
1	Claude Fable 5.1 (max with fallback)	Anthropic	56.8	71	€8.6 / €43.02	1.0M	State-of-the-art reasoning and complex coding tasks requiring maximum intelligence
2	GPT-6 Astra (max)	OpenAI	54.7	Hardware dependent	€8.6 / €43.02	1.0M	High-end proprietary reasoning and large-scale enterprise applications
3	Claude Opus 5 (max)	Anthropic	54.1	52	€4.3 / €21.51	1.0M	Balanced high-intelligence tasks with cost-effective proprietary pricing
4	Claude Fable 5.1 (xhigh with fallback)	Anthropic	53.8	63	€8.6 / €43.02	1.0M	Extreme-effort reasoning tasks requiring maximum logical precision
5	Claude Opus 5 (Adaptive Reasoning, Xhigh Effort)	Anthropic	53.4	53	€4.3 / €21.51	—	High-effort adaptive reasoning for complex problem solving without context needs
6	Claude Fable 5 (with fallback)	Anthropic	53.2	66	€8.6 / €43.02	1.0M	High-intelligence reasoning with robust fallback capabilities
7	Muse Spark 1.3 (max)	Meta	52.7	177	€1.08 / €3.66	1.0M	Ultra-fast high-intelligence proprietary workloads
8	Claude Opus 5 (Adaptive Reasoning, High Effort)	Anthropic	52.0	50	€4.3 / €21.51	—	High-effort adaptive reasoning for complex logic with zero context
9	Claude Fable 5.1 (high with fallback)	Anthropic	51.7	52	€8.6 / €43.02	1.0M	High-effort reasoning with robust fallback and large context
10	GPT-5.6 Sol (max)	OpenAI	51.3	79	€3.44 / €17.21	1.0M	Fast, high-intelligence proprietary tasks with balanced pricing

#1 · proprietary

Claude Fable 5.1 (max with fallback)

Anthropic

Intel: 56.8
Speed: 71 t/s
Cost: €8.6 / €43.02
Ctx: 1.0M

State-of-the-art reasoning and complex coding tasks requiring maximum intelligence

#2 · proprietary

GPT-6 Astra (max)

OpenAI

Intel: 54.7
Cost: €8.6 / €43.02
Ctx: 1.0M

High-end proprietary reasoning and large-scale enterprise applications

#3 · proprietary

Claude Opus 5 (max)

Anthropic

Intel: 54.1
Speed: 52 t/s
Cost: €4.3 / €21.51
Ctx: 1.0M

Balanced high-intelligence tasks with cost-effective proprietary pricing

#4 · proprietary

Claude Fable 5.1 (xhigh with fallback)

Anthropic

Intel: 53.8
Speed: 63 t/s
Cost: €8.6 / €43.02
Ctx: 1.0M

Extreme-effort reasoning tasks requiring maximum logical precision

#5 · proprietary

Claude Opus 5 (Adaptive Reasoning, Xhigh Effort)

Anthropic

Intel: 53.4
Speed: 53 t/s
Cost: €4.3 / €21.51
Ctx: —

High-effort adaptive reasoning for complex problem solving without context needs

#6 · proprietary

Claude Fable 5 (with fallback)

Anthropic

Intel: 53.2
Speed: 66 t/s
Cost: €8.6 / €43.02
Ctx: 1.0M

High-intelligence reasoning with robust fallback capabilities

#7 · proprietary

Muse Spark 1.3 (max)

Claude Opus 5 (Adaptive Reasoning, High Effort)

Anthropic

Intel: 52.0
Speed: 50 t/s
Cost: €4.3 / €21.51
Ctx: —

High-effort adaptive reasoning for complex logic with zero context

#9 · proprietary

Claude Fable 5.1 (high with fallback)

Anthropic

Intel: 51.7
Speed: 52 t/s
Cost: €8.6 / €43.02
Ctx: 1.0M

High-effort reasoning with robust fallback and large context

#10 · proprietary

GPT-5.6 Sol (max)

OpenAI

Intel: 51.3
Speed: 79 t/s
Cost: €3.44 / €17.21
Ctx: 1.0M

Fast, high-intelligence proprietary tasks with balanced pricing

high volume customer support chat
- Blazing speed of 429 tok/s minimizes user wait times
- Highly economical pricing at €0.65/€3.23 per Mtok
- Strong intelligence index of 47.1 for a value-tier model
Gemini 3.8 Flash (high)
complex software engineering and coding
- Highest intelligence index of 56.8 ensures superior logic
- Generous 1,000,000 token context window handles large codebases
- Fast 71 tok/s speed for a top-tier proprietary model
Claude Fable 5.1 (max with fallback)
large scale document data extraction
- Massive 1,048,576 token context window fits entire archives
- High intelligence index of 50.2 ensures accurate data parsing
- Open-source model with 2800B parameters for deep reasoning
Kimi K3 (max)
deep analysis of long documents
- Massive 1,048,576 token context window handles huge documents
- Extremely fast speed of 226 tok/s accelerates analysis
- Cost-effective value tier pricing at €1.08/€3.66 per Mtok
Muse Spark 1.2 (xhigh)
sovereign on premise model deployment
- Top-performing open-source model with 50.2 intelligence index
- Massive 1,048,576 token context window for local data processing
- Deployable on-premise under the Kimi K3 License
Kimi K3 (max)
advanced visual and multimodal processing
- Unmatched intelligence index of 56.8 for complex visual tasks
- Spacious 1,000,000 token context window for multi-image inputs
- Reliable proprietary model from Anthropic with 71 tok/s speed
Claude Fable 5.1 (max with fallback)
autonomous agents and tool use
- Peak intelligence index of 56.8 ensures reliable tool use
- 1,000,000 token context window supports long agent trajectories
- Fast 71 tok/s speed enables rapid multi-step execution
Claude Fable 5.1 (max with fallback)
high accuracy multilingual translation
- Strong 46.9 intelligence index from Alibaba for nuanced translation
- Spacious 1,000,000 token context window for long documents
- Value tier pricing at €1.72/€5.16 per Mtok reduces localization costs
Qwen3.8 Max
high speed structured data extraction
- Massive 1,048,576 token context window captures all target fields
- Blazing speed of 226 tok/s ensures rapid batch processing
- Value tier pricing at €1.08/€3.66 per Mtok keeps extraction cheap
Muse Spark 1.2 (xhigh)

How we rank

Snapshot taken 9/5/2026, 2:01:14 AM
Value score = (intelligence − 25)² × min(1, speed ÷ 30) ÷ (input cost + output cost × 3). Output is weighted ×3 because real usage is output-heavy; intelligence is squared so cheap-but-weak models can't dominate.
Intelligence, price and speed data by Artificial Analysis.

Daily refresh at 04:00 Europe/Amsterdam. Numbers move; we don't smooth them.

What is the best AI model right now?

What is the best open-source AI model?

Which AI model has the best price-to-performance?

Which AI model has the largest context window?

How often is the leaderboard updated?

How are the AI models ranked?

Is the SevenLab AI leaderboard free and independent?

A new model just dropped?We track every major release in our daily AI news feed.Read the latest AI news

We pick the model. Then we build the thing.

Building or buying with AI?

Talk directly with our AI specialists

15 min, no strings

No sales pressure

Prototype in 7 days

The AI Leaderboard

Claude Fable 5.1 (max with fallback)

Kimi K3 (max)

GPT-5.6 Luna (max)

Top 10, side by side

Claude Fable 5.1 (max with fallback)

GPT-6 Astra (max)

Claude Opus 5 (max)

Claude Fable 5.1 (xhigh with fallback)

Claude Opus 5 (Adaptive Reasoning, Xhigh Effort)

Claude Fable 5 (with fallback)

Muse Spark 1.3 (max)

Claude Opus 5 (Adaptive Reasoning, High Effort)

Claude Fable 5.1 (high with fallback)

GPT-5.6 Sol (max)

Which model for which job

high volume customer support chat

complex software engineering and coding

large scale document data extraction

deep analysis of long documents

sovereign on premise model deployment

advanced visual and multimodal processing

autonomous agents and tool use

high accuracy multilingual translation

high speed structured data extraction

How we rank

AI leaderboard FAQ

We pick the model. Then we build the thing.

The AI Leaderboard

Claude Fable 5.1 (max with fallback)

Kimi K3 (max)

GPT-5.6 Luna (max)

◈ Top 10, side by side

high volume customer support chat

complex software engineering and coding

large scale document data extraction

deep analysis of long documents

sovereign on premise model deployment

advanced visual and multimodal processing

autonomous agents and tool use

high accuracy multilingual translation

high speed structured data extraction

▣ How we rank

AI leaderboard FAQ

We pick the model. Then we build the thing.

Top 10, side by side

How we rank