+31 20 705 1010
EnglishNederlandsFrançaisDeutschEspañol
AI · LEADERBOARD

The AI Leaderboard

Compare AI models using a daily ranking based on public benchmarks, speed and cost.

As of 9 November 2026, the top-ranked AI model is Gemini 4 Carbon (high) by Google. The strongest open-source model is Kimi K3.1 Pro, and Claude Haiku 5.5 (max) offers the best price-to-performance. Rankings refresh every day at 04:00 Europe/Amsterdam.

Refreshed09 Nov, 04:01 CETDaily, 04:00 Europe/Amsterdam
Best closed model

Gemini 4 Carbon (high)

Google

Intel
60.0
Speed
105t/s
Cost in/out
€3.58 / €17.91

Evaluation of complex multi-step reasoning tasks requiring high intelligence scores

Best open model
Ki

Kimi K3.1 Pro

Kimi

Intel
46.5
Speed
29t/s
Cost in/out
Provider€1.79 / €8.96

Testing open-weights deployments with a 1,000,000 token context window

Best value91.84

Claude Haiku 5.5 (max)

Anthropic

Intel
36.5
Speed
233t/s
Cost in/out
€0.09 / €0.45

High-frequency conversational interactions requiring rapid output generation

Top 10, side by side

#↑ModelVendorIntelligenceSpeed (tok/s)Cost (€/Mtok)ContextBest for
1
Gemini 4 Carbon (high)
Google60.0105€3.58 / €17.911.0MEvaluation of complex multi-step reasoning tasks requiring high intelligence scores
2
Claude Opus 5.5 (max)
Anthropic59.195€3.58 / €17.911.0MTesting on demanding analytical prompts within a 1,000,000 token context window
3
Claude Opus 5.5 (Xhigh, Default Fallback)
Anthropic57.483€3.58 / €17.91—Evaluating high-intelligence proprietary workloads where context limit is unspecified
4
GPT-6 Astra (max)
OpenAI56.956€8.96 / €44.781.0MTesting deep reasoning scenarios where generation speed is not the primary factor
5
GPT-6.1 Sol (max)
OpenAI54.685€1.79 / €8.961.0MBenchmarking high-intelligence queries requiring balanced throughput and pricing
6
Claude Sonnet 5.5 (max)
Anthropic54.5136€1.79 / €8.961.0MHigh-throughput tasks requiring a 1,000,000 token context at 136 tokens per second
7
Claude Opus 5.5 (High, Default Fallback)
Anthropic51.878€3.58 / €17.91—Assessing balanced reasoning tasks on a proprietary configuration
8
GPT-6.1 Terra (max)
OpenAI51.3115€0.9 / €4.481.0MTesting high-volume requests needing moderate intelligence at €0.90 per Mtok input
9
Gemini 4 Argon (high)
Google50.492€3.58 / €17.911.0MTesting long-context proprietary processing at 92 tokens per second
10
Muse Wonder Full (max)
Meta50.1Hardware dependentPricing TBA1.0MEvaluating zero-cost API tiers across extensive 1,000,000 token inputs
#1 · proprietary

Gemini 4 Carbon (high)

Google

Intel
60.0
Speed
105 t/s
Cost
€3.58 / €17.91
Ctx
1.0M

Evaluation of complex multi-step reasoning tasks requiring high intelligence scores

#2 · proprietary

Claude Opus 5.5 (max)

Anthropic

Intel
59.1
Speed
95 t/s
Cost
€3.58 / €17.91
Ctx
1.0M

Testing on demanding analytical prompts within a 1,000,000 token context window

#3 · proprietary

Claude Opus 5.5 (Xhigh, Default Fallback)

Anthropic

Intel
57.4
Speed
83 t/s
Cost
€3.58 / €17.91
Ctx
—

Evaluating high-intelligence proprietary workloads where context limit is unspecified

#4 · proprietary

GPT-6 Astra (max)

OpenAI

Intel
56.9
Speed
56 t/s
Cost
€8.96 / €44.78
Ctx
1.0M

Testing deep reasoning scenarios where generation speed is not the primary factor

#5 · proprietary

GPT-6.1 Sol (max)

OpenAI

Intel
54.6
Speed
85 t/s
Cost
€1.79 / €8.96
Ctx
1.0M

Benchmarking high-intelligence queries requiring balanced throughput and pricing

#6 · proprietary

Claude Sonnet 5.5 (max)

Anthropic

Intel
54.5
Speed
136 t/s
Cost
€1.79 / €8.96
Ctx
1.0M

High-throughput tasks requiring a 1,000,000 token context at 136 tokens per second

#7 · proprietary

Claude Opus 5.5 (High, Default Fallback)

Anthropic

Intel
51.8
Speed
78 t/s
Cost
€3.58 / €17.91
Ctx
—

Assessing balanced reasoning tasks on a proprietary configuration

#8 · proprietary

GPT-6.1 Terra (max)

OpenAI

Intel
51.3
Speed
115 t/s
Cost
€0.9 / €4.48
Ctx
1.0M

Testing high-volume requests needing moderate intelligence at €0.90 per Mtok input

#9 · proprietary

Gemini 4 Argon (high)

Google

Intel
50.4
Speed
92 t/s
Cost
€3.58 / €17.91
Ctx
1.0M

Testing long-context proprietary processing at 92 tokens per second

#10 · proprietary

Muse Wonder Full (max)

Meta

Intel
50.1
Cost
Pricing TBA
Ctx
1.0M

Evaluating zero-cost API tiers across extensive 1,000,000 token inputs

Which model for which job

Model recommendations for nine tasks our teams work on.

  1. Customer support routing and chat

    • Offers 233 tok/s generation speed for responsive chat interfaces
    • Priced at €0.09 input and €0.45 output per Mtok to manage high volume
    • Includes a 1,000,000 token context window for extensive transcripts
    Claude Haiku 5.5 (max)

  2. Code generation and synthesis

    • Highest intelligence index of 60 among all evaluated models
    • Outputs at 105 tok/s for rapid syntax generation
    • 1,000,000 token context window allows large repository ingestion
    Gemini 4 Carbon (high)

  3. Large document data extraction

    • Offers 1,048,576 tokens of context capacity for expansive batches
    • Generation speed of 147 tok/s ensures swift batch processing
    • Context retention across large text requires empirical testing
    Muse Wonder (max)

  4. Long-context record analysis

    • Context capacity reaches 1,048,576 tokens for deep record analysis
    • Runs at 147 tok/s to reduce processing latency on long inputs
    • Recall consistency across full context must be verified in tests
    Muse Wonder (max)

  5. On-premises open-weights deployment

    • Highest intelligence index of 46.5 in the open-weights category
    • 1,000,000 token context window available for self-hosted instances
    • Self-hosting hardware demands depend on unlisted parameter count
    Kimi K3.1 Pro

  6. Multimodal visual reasoning

    • Highest comparative intelligence score of 60 among listed candidates
    • Large 1,000,000 token context window accepts large image sequences
    • Image parsing accuracy requires task-specific empirical validation
    Gemini 4 Carbon (high)

  7. Autonomous agentic workflow execution

    • Top intelligence index of 60 supports multi-step instruction planning
    • 1,000,000 token context maintains extensive execution histories
    • Tool call reliability must be tested directly in practical workflows
    Gemini 4 Carbon (high)

  8. Multilingual document translation

    • Leading intelligence score of 60 aids nuanced linguistic translation
    • Generates 105 tok/s to support high-throughput translation jobs
    • Specific language pair nuance requires targeted bench testing
    Gemini 4 Carbon (high)

  9. Structured output and schema extraction

    • Extensive 1,048,576 token context supports massive raw document sets
    • Operates at 147 tok/s to minimise parsing runtimes
    • Schema syntax validity must be verified through format test suites
    Muse Wonder (max)

How we rank

  • Snapshot taken 09 Nov, 04:01 CET
  • 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.

AI leaderboard FAQ

Want this in your company?

The leaderboard shows which model scores best. We build the AI agents and software around it, with a working POC on your data in 7 days.

Automate with AI agentsCustom AI apps and platforms
A new model just dropped?We track every major release in our daily AI news feed.Read the latest AI news
Explore more

Related AI topics

AI quality control for product inspection | SevenLab

From model selection to a working application.

Building or buying with AI?

Talk directly with our AI specialists

15 minutes, no obligation
No sales pressure
Prototype in 7 days