Intelligence index
Highest available values
The source’s asterisk is preserved beside the score.
Explore model benchmarksChoose your model
Choose your model
with data in mind.
Compare intelligence, cost and speed. Explore how models and evaluation configurations differ for your use case.
API benchmarks do not describe consumer subscription features or prices.
Several perspectives at a glance
Intelligence, cost or speed?
Each chart shows a separate metric. Table filters also apply to the charts, and rankings cover only the entries in this snapshot.
Model comparison charts
score
Cost per task
Costs for entries with the highest intelligence scores
Source values are rounded. $0.00 does not mean free use.
Output speed
Highest available values
Output speed is different from response-start latency.
Detailed comparison
Compare models side by side
The full row name, including configurations in parentheses, identifies its evaluation.
272 of 272 entries
Select 2–4 entries for a side-by-side comparison. Your selections remain when you change filters.
| Select for comparison | score | USD / task | tokens / second | seconds | seconds | tokens | Source | |
|---|---|---|---|---|---|---|---|---|
| Claude Opus 5.5 (max with fallback)Anthropic | 58 | $5.98 | 96 | 638.47 | 643.67 | 1M | Details | |
| Claude Sonnet 5.5 (max with fallback)Anthropic | 56 | $5.46 | 134 | 363.89 | 367.63 | 1M | Details | |
| Claude Opus 5.5 (xhigh with fallback)Anthropic | 56 | $3.46 | 82 | 109.13 | 115.22 | 1M | Details | |
| Muse Wonder Full (max)Meta | 56 | Data unavailable | Data unavailable | Data unavailable | Data unavailable | 1M | Details | |
| Kimi K3.1 ProKimi | 55 | $2.11 | 58 | 3.07 | 46.53 | 1M | Details | |
| Claude Opus 5.5 (high with fallback)Anthropic | 54 | $1.82 | 80 | 27.36 | 33.58 | 1M | Details | |
| Claude Fable 5.1 (max with fallback)Anthropic | 53 | $7.63 | 70 | 166.61 | 173.79 | 1M | Details | |
| Claude Fable 5.1 (xhigh with fallback)Anthropic | 53 | $5.98 | 63 | 63.44 | 71.41 | 1M | Details | |
| GPT-6 Astra (max)OpenAI | 53 | $3.26 | 68 | 272.88 | 280.26 | 1M | Details | |
| Gemini 4 Argon (high)Google | 53 | $1.99 | 79 | 57.01 | 63.35 | 1M | Details | |
| Muse Wonder (max)Meta | 52 | $4.54 | 162 | 25.23 | 40.7 | 1.05M | Details | |
| GPT-6 Astra (xhigh)OpenAI | 52 | $2.31 | 61 | 133.2 | 141.33 | 1M | Details |
Claude Opus 5.5 (max with fallback)
Anthropic
- Intelligence indexscore
- 58
- Cost per taskUSD / task
- $5.98
- Output speedtokens / second
- 96
- First-chunk latencyseconds
- 638.47
- Total response timeseconds
- 643.67
- Context windowtokens
- 1M
Claude Sonnet 5.5 (max with fallback)
Anthropic
- Intelligence indexscore
- 56
- Cost per taskUSD / task
- $5.46
- Output speedtokens / second
- 134
- First-chunk latencyseconds
- 363.89
- Total response timeseconds
- 367.63
- Context windowtokens
- 1M
Claude Opus 5.5 (xhigh with fallback)
Anthropic
- Intelligence indexscore
- 56
- Cost per taskUSD / task
- $3.46
- Output speedtokens / second
- 82
- First-chunk latencyseconds
- 109.13
- Total response timeseconds
- 115.22
- Context windowtokens
- 1M
Muse Wonder Full (max)
Meta
- Intelligence indexscore
- 56
- Cost per taskUSD / task
- Data unavailable
- Output speedtokens / second
- Data unavailable
- First-chunk latencyseconds
- Data unavailable
- Total response timeseconds
- Data unavailable
- Context windowtokens
- 1M
Kimi K3.1 Pro
Kimi
- Intelligence indexscore
- 55
- Cost per taskUSD / task
- $2.11
- Output speedtokens / second
- 58
- First-chunk latencyseconds
- 3.07
- Total response timeseconds
- 46.53
- Context windowtokens
- 1M
Claude Opus 5.5 (high with fallback)
Anthropic
- Intelligence indexscore
- 54
- Cost per taskUSD / task
- $1.82
- Output speedtokens / second
- 80
- First-chunk latencyseconds
- 27.36
- Total response timeseconds
- 33.58
- Context windowtokens
- 1M
Claude Fable 5.1 (max with fallback)
Anthropic
- Intelligence indexscore
- 53
- Cost per taskUSD / task
- $7.63
- Output speedtokens / second
- 70
- First-chunk latencyseconds
- 166.61
- Total response timeseconds
- 173.79
- Context windowtokens
- 1M
Claude Fable 5.1 (xhigh with fallback)
Anthropic
- Intelligence indexscore
- 53
- Cost per taskUSD / task
- $5.98
- Output speedtokens / second
- 63
- First-chunk latencyseconds
- 63.44
- Total response timeseconds
- 71.41
- Context windowtokens
- 1M
GPT-6 Astra (max)
OpenAI
- Intelligence indexscore
- 53
- Cost per taskUSD / task
- $3.26
- Output speedtokens / second
- 68
- First-chunk latencyseconds
- 272.88
- Total response timeseconds
- 280.26
- Context windowtokens
- 1M
Gemini 4 Argon (high)
- Intelligence indexscore
- 53
- Cost per taskUSD / task
- $1.99
- Output speedtokens / second
- 79
- First-chunk latencyseconds
- 57.01
- Total response timeseconds
- 63.35
- Context windowtokens
- 1M
Muse Wonder (max)
Meta
- Intelligence indexscore
- 52
- Cost per taskUSD / task
- $4.54
- Output speedtokens / second
- 162
- First-chunk latencyseconds
- 25.23
- Total response timeseconds
- 40.7
- Context windowtokens
- 1.05M
GPT-6 Astra (xhigh)
OpenAI
- Intelligence indexscore
- 52
- Cost per taskUSD / task
- $2.31
- Output speedtokens / second
- 61
- First-chunk latencyseconds
- 133.2
- Total response timeseconds
- 141.33
- Context windowtokens
- 1M
Showing 1 to 12 of 272
1 / 23
Unavailable does not mean zero. Asterisks and numeric precision are preserved from the source. Context retains the source’s k/M labels; costs are US dollars per evaluation task.
Clear data for informed decisions
From the source to this page
These figures are not ElmosGPT test results. The source’s public table was captured at a specific time, then restructured only for consistent display and sorting.
Artificial Analysis methodology- 1
Capture the public data
272 visible rows were saved from the public table with their full names, creators and model links. Retrieved: 25 October 2026 at 13:44 (Tehran time).
- 2
Preserve meaning and missing values
Missing metrics remain null. Asterisks and rounded figures are retained. The k and M context labels are converted to token counts only for sorting.
- 3
Compare within the evaluation’s scope
Review the task, model configuration and access terms separately. Use the source links for the live table, calculation methodology and provider details.
Intelligence indexscore
The Artificial Analysis Intelligence Index in the reference table. A higher score is better within this evaluation.
Cost per taskUSD / task
Cost per task in the source evaluation, in US dollars. It is not a subscription price or a price per million tokens. Source figures are rounded.
Output speedtokens / second
Median output token speed in the reference table. A higher value means faster generation.
First-chunk latencyseconds
The source’s Latency First Chunk: time until the first response chunk, which differs from the total response duration.
Total response timeseconds
Total Response in the reference table. A lower value means a shorter duration under that evaluation’s conditions.
Context windowtokens
The combined input and output limit stated in the table. A model’s output limit may be smaller.
Common questions
Before drawing conclusions
Are these ChatGPT or Claude subscription prices?
No. The cost column shows US dollars per task in the source’s evaluation. It is not a consumer subscription price, a price per million tokens or an ElmosGPT product price. Review consumer plan features and usage limits separately.
Why do some cells say “Data unavailable”?
The public source table did not display a value for that model and metric when this snapshot was retrieved. ElmosGPT does not estimate it or turn it into zero. Missing values remain at the end in both ascending and descending order.
Does a higher score always mean a better choice?
A score is one part of the decision. Consider the task, response time, cost, context window and practical limits together. Different evaluation configurations of the same model can also produce different results.
What do an asterisk and a $0.00 cost mean?
The asterisk is preserved from the source table. Visit the model’s source page for its precise meaning. Costs retain the source’s displayed precision: $0.00 is a rounded value and does not by itself mean free use.
Is this live data, and did ElmosGPT rerun the tests?
This page uses a saved snapshot of the public Artificial Analysis table. The retrieval time is shown above; opening this page does not fetch live data. ElmosGPT checked the structure and value types but did not rerun the source’s tests.
Why does one model family have several rows?
The source lists different modes and evaluation configurations separately. The full row name, including text in parentheses, identifies that entry. The number of rows is not necessarily the number of distinct model families.
Connect the comparison to your task.
Read the ElmosGPT choosing guide to understand the differences between tools, models and consumer plans.
Data source: Artificial Analysis · Saved snapshot 25 October 2026 at 13:44 (Tehran time)ElmosGPT is not affiliated with or endorsed by the source.