Skip to content
Mindverse Logo

// As of October 25, 2026

AI models compared

GPT, Claude, Gemini, DeepSeek and other leading language models side by side: 27 models by capability, price and context window. Every figure has a source and a date, so you can check it.

Jump to the comparison Try the models in Studio
Models
27 compared
Index
Epoch AI (ECI)
In Studio
13 models

Top of the Epoch Capabilities Index

  1. 1 Claude Opus 5.5 Anthropic 168.8 165–174
  2. 2 GPT-6 Astra OpenAI 168.8 165–175
  3. 3 GPT-6.1 Sol OpenAI 168.4 165–174
  4. 4 Claude Fable 5.1 Anthropic 166.5 163–171
  5. 5 Gemini 4 Argon Google 166.5 163–172

The uncertainty ranges at the top overlap, so the order of the leading models is close. Source: Epoch AI, CC BY 4.0.

The picks at a glance

Highest capability

Claude Opus 5.5

Anthropic

ECI 168.8 · range 165–174

A tight race: the uncertainty ranges of 7 other models overlap with this one.

Top tier at the best price

GPT-6.1 Sol

OpenAI

ECI 168.4 for $4.00 per 1M tokens

Cheapest model within 5 points of the top.

GPT-5 level for the least money

Claude Haiku 5.5

Anthropic

ECI 159.3 for $0.20 per 1M tokens

Cheapest model with at least 150 points, the score of GPT-5.

Strongest open-weight model

Kimi K3

Moonshot AI

ECI 157.9

Weights published under the Kimi K3 License. Can be self-hosted.

// Capability and price

More capability does not always cost more

Each dot is a model and the vertical line its uncertainty range. Strong, cheap models sit top left. The dashed line connects the models no other model beats at their price.

Claude Opus 5.5 (Anthropic): ECI 168.8, $8.00 per 1M tokens Claude Opus 5.5 GPT-6 Astra (OpenAI): ECI 168.8, $20 per 1M tokens GPT-6.1 Sol (OpenAI): ECI 168.4, $4.00 per 1M tokens GPT-6.1 Sol Claude Fable 5.1 (Anthropic): ECI 166.5, $20 per 1M tokens Gemini 4 Argon (Google): ECI 166.5, $4.00 per 1M tokens Claude Sonnet 5.5 (Anthropic): ECI 166.2, $4.00 per 1M tokens GPT-6 Sol (OpenAI): ECI 163.7, $4.00 per 1M tokens Claude Haiku 5.5 (Anthropic): ECI 159.3, $0.20 per 1M tokens Claude Haiku 5.5 Kimi K3 (Moonshot AI): ECI 157.9, $6.00 per 1M tokens Gemini 3.8 Flash (Google): ECI 157.4, $1.50 per 1M tokens Qwen 3.8 Max (Alibaba): ECI 157.4, $3.00 per 1M tokens GPT-6 Luna (OpenAI): ECI 156.8, $0.20 per 1M tokens DeepSeek V4.1 Flash (DeepSeek): ECI 155.9, $0.53 per 1M tokens DeepSeek V4 Pro (DeepSeek): ECI 155.6, $1.98 per 1M tokens Gemini 3.1 Pro (Google): ECI 155.3, $4.50 per 1M tokens Grok 4.7 (xAI): ECI 154.4, $3.00 per 1M tokens GPT-5.2 (OpenAI): ECI 153.7, $4.81 per 1M tokens GLM-5.3 Flash (Z.ai): ECI 153.1, $0.24 per 1M tokens Claude Sonnet 4.6 (Anthropic): ECI 152.8, $6.00 per 1M tokens Mistral Large 4 (Mistral AI): ECI 150.0, $2.07 per 1M tokens MiniMax M3 (MiniMax): ECI 146.9, $0.53 per 1M tokens GPT-5 mini (OpenAI): ECI 145.3, $0.69 per 1M tokens Claude Haiku 4.5 (Anthropic): ECI 142.2, $2.00 per 1M tokens Mistral Medium 3.5 (Mistral AI): ECI 141.4, $3.00 per 1M tokens
Proprietary Open Weights In Mindverse Studio Uncertainty range Efficiency frontier Capability: Epoch Capabilities Index (Epoch AI, CC BY 4.0). Price: provider list prices, input to output weighted 3:1. Models without a list price are missing from the chart but listed in the table. As of October 25, 2026.

// All figures

Every model in detail

Click a column heading to sort. The filter above applies here too.

AI models compared: capability, price and context window
Claude Opus 5.5 Anthropic 168.8 165–174 $8.00 $4.00 / $20 1M Sep 2026
GPT-6 Astra OpenAI Price up to 272K input tokens; higher above. 168.8 165–175 $20 $10 / $50 1.05M Sep 2026
GPT-6.1 Sol OpenAI Price up to 272K input tokens; higher above. 168.4 165–174 $4.00 $2.00 / $10 1.05M Sep 2026
Claude Fable 5.1 Anthropic 166.5 163–171 $20 $10 / $50 1M Sep 2026
Gemini 4 Argon Google Price until 30 Nov 2026; from 1 Dec 2026: $4 / $20. 166.5 163–172 $4.00 $2.00 / $10 1.05M Sep 2026
Claude Sonnet 5.5 Anthropic In Mindverse Studio 166.2 163–171 $4.00 $2.00 / $10 1M Sep 2026
GPT-6 Sol OpenAI In Mindverse Studio Successor: GPT-6.1 Sol. Price up to 272K input tokens. 163.7 161–167 $4.00 $2.00 / $10 1.05M Sep 2026
Muse Wonder Meta No public API list price. 161.9 159–165 n/a n/a Oct 2026
Claude Haiku 5.5 Anthropic Price up to 100,000 prompt tokens; higher above. 159.3 156–162 $0.20 $0.10 / $0.50 1M Oct 2026
Kimi K3 Moonshot AI · Kimi K3 License Open Weights 157.9 155–161 $6.00 $3.00 / $15 1.05M Jul 2026
Gemini 3.8 Flash Google In Mindverse Studio Price until 31 Dec 2026; from 1 Jan 2027: $1.50 / $7.50. 157.4 155–161 $1.50 $0.75 / $3.75 1.05M Sep 2026
Qwen 3.8 Max Alibaba Price for the international region. 157.4 155–160 $3.00 $2.00 / $6.00 n/a Aug 2026
GPT-6 Luna OpenAI In Mindverse Studio Price up to 272K input tokens; higher above. 156.8 154–159 $0.20 $0.10 / $0.50 1.05M Sep 2026
DeepSeek V4.1 Flash DeepSeek · MIT In Mindverse Studio Runs in Germany Open Weights Peak price, half off-peak. ECI for the September version; the API has served version 1017 since October. 155.9 154–159 $0.53 $0.30 / $1.20 1M Sep 2026
DeepSeek V4 Pro DeepSeek · MIT In Mindverse Studio Runs in Germany Open Weights Version 0813. Peak price, half off-peak. 155.6 154–158 $1.98 $1.32 / $3.96 1M Aug 2026
Gemini 3.1 Pro Google In Mindverse Studio A preview at Google. Price up to 200K prompt tokens. 155.3 153–158 $4.50 $2.00 / $12 1.05M Feb 2026
Gemini 3.5 Flash Google In Mindverse Studio No longer on Google's price list; Google routes API requests to Gemini 3.6 Flash. 154.9 153–157 n/a n/a May 2026
Grok 4.7 xAI Price below 200K prompt tokens. 154.4 153–157 $3.00 $2.00 / $6.00 500K Sep 2026
GPT-5.2 OpenAI In Mindverse Studio 153.7 152–155 $4.81 $1.75 / $14 400K Dec 2025
GLM-5.3 Flash Z.ai · MIT Open Weights 153.1 151–156 $0.24 $0.15 / $0.50 n/a Aug 2026
Claude Sonnet 4.6 Anthropic In Mindverse Studio A legacy model at Anthropic. 152.8 150–155 $6.00 $3.00 / $15 1M Feb 2026
Mistral Large 4 Mistral AI Public preview. Weights not published yet. 150.0 147–153 $2.07 $1.36 / $4.18 1M Oct 2026
MiniMax M3 MiniMax · MiniMax Community License Open Weights Price up to 512K input tokens. 146.9 142–150 $0.53 $0.30 / $1.20 n/a Jun 2026
GPT-5 mini OpenAI In Mindverse Studio Listed as deprecated by OpenAI. 145.3 143–147 $0.69 $0.25 / $2.00 400K Aug 2025
Claude Haiku 4.5 Anthropic In Mindverse Studio A legacy model; successor: Claude Haiku 5.5. 142.2 139–144 $2.00 $1.00 / $5.00 200K Oct 2025
Mistral Medium 3.5 Mistral AI · Modified MIT Open Weights 141.4 139–144 $3.00 $1.50 / $7.50 256K Apr 2026
GPT OSS 120B OpenAI · Apache 2.0 In Mindverse Studio Runs in Germany Open Weights No vendor price: open weights, so the price depends on the host. 139.8 136–142 n/a 131K Aug 2025

Price: provider list prices, input and output weighted 3:1; below it input / output. Bar: ECI with Epoch AI’s uncertainty range. n/a = not published.

// By task

Which model for which job?

Each pick follows a fixed rule applied to the data above. The rule is shown, so you can weigh it against your own priorities.

// Data protection

Where a model runs matters too

Whether a deployment is GDPR-compliant depends on processing, contract and location, not on the benchmark. This is how the models run in Mindverse Studio:

On infrastructure in Germany

DeepSeek V4.1 Flash, DeepSeek V4 Pro and GPT OSS 120B run as sovereign models entirely on Mindverse infrastructure in Germany.

Through external providers

GPT, Claude and Gemini models run with external providers. Your data processing agreement lists the subprocessors involved. For especially confidential data, choose a sovereign model.

A dedicated instance

Enterprise customers can get a dedicated LLM instance, physically and logically isolated from other customer environments.

Security and GDPR at Mindverse Discuss an enterprise instance

// Hydra · in development

Does it always have to be the strongest model?

For many tasks a cheaper model is enough. Hydra, our routing project, picks one model per request based on task, quality, cost and response time. Hydra is still in development and its benchmarks are internal measurements, which is why they are not in this table.

Explore Hydra

// Method

How to read the numbers

Epoch Capabilities Index
The Epoch Capabilities Index (ECI) by Epoch AI combines results from more than 50 benchmarks into one shared scale, anchored so that Claude 3.5 Sonnet scores 130 and GPT-5 scores 150. Absolute values mean nothing on their own; the gaps between models do. Epoch AI gives an uncertainty range for every score.
Price
Provider list prices in US dollars per 1M tokens, standard tier. For the comparison we weight input and output 3:1, a common ratio for chat applications. Long-prompt tiers, cache discounts and batch prices are not included; caveats are noted with each model.
Context
Maximum context length per the provider, in tokens. In Mindverse Studio the usable context window can differ by model.
What is missing
We do not show speed or specialist benchmarks such as coding: for most models here there are no openly licensed, comparable measurements. For more leaderboards, see Artificial Analysis and Arena .
Limits
Benchmarks measure averages over standardised tasks. What counts for your work is a test with your own data: in Mindverse Studio you can try the models side by side.

// FAQ

Frequently asked questions about AI models

Which AI model is the best right now?
On the Epoch Capabilities Index (as of October 25, 2026), Claude Opus 5.5 by Anthropic leads with 168.8 points, just ahead of GPT-6 Astra (168.8) and GPT-6.1 Sol (168.4). The gaps at the top are smaller than the uncertainty ranges Epoch AI reports, so several models are effectively tied. The best model for you therefore depends more on task, price and data protection. The recommendations above state the rule behind each pick.
What does the Epoch Capabilities Index measure?
The Epoch Capabilities Index (ECI) by Epoch AI combines results from more than 50 benchmarks into one shared scale, anchored so that Claude 3.5 Sonnet scores 130 and GPT-5 scores 150. Absolute values mean nothing on their own; the gaps between models do. Epoch AI gives an uncertainty range for every score.
Which AI models can I use in a GDPR-compliant way?
GDPR compliance is not a property of the model but of the processing: location, data processing agreement and purpose limitation decide. In Mindverse Studio, DeepSeek V4.1 Flash, DeepSeek V4 Pro and GPT OSS 120B run as sovereign models on infrastructure in Germany. Mindverse signs a data processing agreement under Art. 28 GDPR with every customer, and your content is never used for training. More on security and GDPR.
How much do large language models cost?
The range is wide: in this comparison the weighted list price runs from $0.20 (Claude Haiku 5.5) to $20 (Claude Fable 5.1) per 1M tokens. Output tokens almost always cost several times more than input, and reasoning models generate extra thinking tokens. In Mindverse Studio you use the models within your plan; see current plans on pricing.
Open-weight or proprietary: which is better?
The strongest proprietary model, Claude Opus 5.5, scores 168.8 points; the strongest open-weight model, Kimi K3, scores 157.9. In exchange, open-weight models can be self-hosted: an advantage when data must stay in your own data centre or costs at high volume need to stay predictable.
Can I use several models in one workspace?
Yes. Mindverse Studio brings GPT, Claude, Gemini and sovereign models such as DeepSeek into one workspace: 13 of the models in this comparison. Use them side by side or let Auto mode pick the right model for each request.

// Compare for yourself

Test the models on your own tasks

Try free for 7 days. GDPR-compliant, hosted in Germany — no credit card required.

Start free trial Book a demo

GDPR-COMPLIANT · SERVERS IN GERMANY · SOC 2-ORIENTED PROCESSES