Artificial Analysis updates the Intelligence Index: Claude Opus 5.5 leads by 58 points

AT A GLANCE

The report says the new version of the Intelligence Index adds real programming and research tasks, with Claude Opus 5.5 taking 58 points to rank first.

Article

AA Major overhaul of the Intelligence Index: Claude takes the lead, while DeepSeek and GLM trail behind 26 points

Beating AI breaking news,AI evaluation firm Artificial Analysis has significantly revised its Intelligence Index leaderboard. The new version adds 132 programming tasks from real open-source projects. The code written by AI must not only pass tests, but also meet the standard that project maintainers would be willing to merge. It also adds 70 research tasks, requiring AI to actually complete the research workflow in a terminal. To reduce the impact of models having seen the questions in advance, the new version's 60%scores are weighted based on private test sets or hidden answers.

Anthropic and OpenAI still dominate the top of the leaderboard. Claude Opus 5.5 scores 58 points to take first place, while GPT-6 Astra and GPT-6.1 Sol score 57 points and 55 points, respectively. Kimi K3.1 Pro scores 46 points, but its weights have not yet been released.

The open-weight DeepSeek V4.1 Flash and GLM-5.3 each score only 32 points, trailing the leader by a full 26 points and even falling below Claude Haiku 5.5 at 36 points.

In the newly added research tasks,GPT-6 Astra achieves a completion rate of 63%，Kimi K3.1 Pro records 34%，DeepSeek V4.1 Flash manages only 13%.

Although GPT-6.1 Sol scores 5.5 points lower than Claude Opus 3, based on publicly listed API prices, the estimated average cost per comprehensive evaluation task is only 1.27 USD, compared with 10.04 USD, nearly 8 times as much.

The new version's questions are also more demanding overall, with the median output Token for each task rising from Token in the previous version to 35000 in the new version.

Original link 59000 https://m.theblockbeats.info/flash/374127

Key points

Artificial Analysis has significantly revised its Intelligence Index leaderboard.
The new version adds 132 programming tasks from real open-source projects and 70 research tasks.
The new version's 60%scores are weighted based on private test sets or hidden answers.
Claude Opus 5.5 scores 58 points,GPT-6 Astra scores 57 points,GPT-6.1 Sol scores 55 points.
Kimi K3.1 Pro scores 46 points; the article says its weights have not yet been released.
DeepSeek V4.1 Flash and GLM-5.3 each score 32 points, trailing the leader by 26 points.
In the newly added research tasks,GPT-6 Astra、Kimi K3.1 Pro、DeepSeek V4.1 Flash have completion rates of 63%、34%and 13%。
Based on publicly listed API prices,GPT-6.1 Sol and Claude Opus 5.5 require, on average, 1.27 USD and 10.04 USD。
The median output Token for each task rises from the previous version's 35000 to the new version's 59000。

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

This leaderboard not only checks whether answers pass tests, but also requires the code to meet the standard that project maintainers would be willing to merge, and has models complete research workflows in a terminal. The reported overall scores, research-task completion rates, task costs, and Token usage respectively reflect different evaluation dimensions and should not be treated as the same metric.

Why it matters to readers

If the report is accurate, the new evaluation places greater emphasis on real programming, research workflows, and hidden tests; the results show clear differences among models in overall scores, research-task completion rates, and publicly listed API cost estimates.

When comparing models, beginners should not look only at total scores; they can also consider task types, completion rates, whether the weights are open,Token consumption, and API costs.

Risks and unknowns

The evidence consists of a media news brief, without the original Artificial Analysis ranking or methodology explanation.
The rankings, model versions, scores, and completion rates stated in the article lack independently verifiable supporting materials.
Public API pricing estimate: the specific calculation methodology is not disclosed.
The evidence does not indicate whether these evaluation results can be replicated in real-world development or scientific research tasks.

Artificial Analysis updates the Intelligence Index: Claude Opus 5.5 leads by 58 points

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center