Opus 5.5 Tops AI Index at 58; GPT-6.1 Sol Costs 8x Less

Artificial Analysis said in a post on X on 28 October 2026 that Claude Opus 5.5 (max) “leads at 58” on version 5 of its Intelligence Index, ahead of GPT-6 Astra (max) at 57 and GPT-6.1 Sol (max) at 55. The benchmarking firm called v5 “our biggest upgrade since v4 launched in January.”

Its leaderboard puts the cost of running the Index at $10.04 per task for Opus 5.5 (max) and $1.27 for GPT-6.1 Sol (max). By our arithmetic, that makes the OpenAI model about 7.9 times cheaper per task, three points behind.

Short explainer · AI Index v5 brief

A vertical news Short narrated from the published headlines and opening lines of this piece and four related FourWeekMBA pieces on AI models and their costs.

Business Pill · Model routing

An evergreen Business Engineer explainer on model routing: sending each request to the cheapest model that can handle it. It explains the general idea only and does not discuss Artificial Analysis or these scores.

What Changed in Version 5

Artificial Analysis says that “with v5, 60% of the Index is now based on evaluations with private test sets or held-out answers.” Its post adds: “For the first time, the majority of the Index is built on private datasets.”

The Index now combines nine evaluations, and the methodology page says it “is calculated as a weighted average across four categories: Knowledge Work (25%), Coding (25%), Science (25%), and General (25%).”

Two benchmarks are new. FrontierCode 2.0 is described as “Cognition’s newly-launched private benchmark of real-world coding tasks, graded on whether a repository maintainer would merge the agent’s change”, and Terminal-Bench-Science adds “70 agentic scientific research tasks, run in a terminal environment”.

Three were upgraded: HLE-Diamond replaces Humanity’s Last Exam, AA-Omniscience moves to v1.1, and CritPt v2 is now “a private, 40-question hard subset”. GDPval-AA v2.1, SciCode and AA-LCR v1.1 were removed from the Index.

What Was Posted on X

Artificial Analysis announced Intelligence Index v5 on X, saying that 60% of the Index is now based on evaluations with private test sets or held-out answers and that Claude Opus 5.5 (max) leads at 58, ahead of GPT-6 Astra (max) at 57 and GPT-6.1 Sol (max) at 55.

Who Leads, and What a Task Costs

According to the post, “Anthropic and OpenAI hold the top five models”, and “Gemini 4 Argon (high) is next at 50”. On the leaderboard, Claude Sonnet 5.5 (max) scores 54 at $10.02 per task, and GPT-6.1 Terra (max) scores 51 at $0.96.

Artificial Analysis writes that “cost per task varies by more than 13x among the leading models”. Among the 10 highest-scoring models it shows, it names GPT-6.1 Terra (max) as the cheapest at $0.96 per task and Claude Fable 5.1 (max) as the most expensive at $13.04.

Speed differs as well. The leaderboard lists a median of 98 tokens per second for Opus 5.5 (max) and 81 for GPT-6.1 Sol (max).

The Open-Weights Gap

The post says that “the open weights gap widens from 3 points to 12”. It names Kimi K3.1 Pro as the top open-weights model at 46 and notes that its weights are not yet released.

Among models whose weights are out, Artificial Analysis says DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) lead at 32, “26 points behind the leader and below Claude Haiku 5.5 (max, 36)”.

Artificial Analysis also launched a Progress Index v1 alongside v5, which it says “tracks model capability over time on one continuous scale.”

Connecting the Coverage

FrontierCode is the Cognition benchmark we covered a day earlier in Opus 5.5 tops Cognition’s test; GPT-6.1 Sol costs 12x less. It now carries 15% of the Index.

Demand for the model shows up elsewhere too: we reported that Anthropic tokens on OpenRouter are up 3.6x since Opus 5.5, and on the same day as this Index, Similarweb counted a record 47M daily visits to Claude’s website.

The cost side links to our report that Nvidia Rubin gets up to 3.2x GB300’s tokens per dollar, a separate measure of what the hardware underneath costs to run.

The Structural Read

As we read it, the move to private test sets matters for how the scores can be used: Artificial Analysis says 60% of the Index now rests on evaluations with private test sets or held-out answers.

By our arithmetic, Opus 5.5 (max) costs about 7.9 times as much per Index task as GPT-6.1 Sol (max), for a score three points higher.

Artificial Analysis says the open-weights gap widened from 3 points to 12, and that the best models with released weights score 32.

Three Implications

FOR ANTHROPIC Claude Opus 5.5 (max) leads the new Index at 58, and a Claude model also holds the most expensive slot among the 10 highest-scoring models, Claude Fable 5.1 (max) at $13.04 per task, per Artificial Analysis.

FOR OPENAI Artificial Analysis lists GPT-6 Astra (max) one point behind at 57, and GPT-6.1 Sol (max) at 55 for $1.27 per task.

FOR BUYERS COMPARING MODELS Artificial Analysis says cost per task varies by more than 13x among the leading models, so the leaderboard shows score and cost side by side.

The Business Engineer Lens

This story maps onto the Business Engineer framework The Open vs Closed Meta-Framework.

The framework starts from the line “Every technology stack has layers.” It sorts those layers into scarce and abundant ones and argues for keeping the scarce layer proprietary while opening the abundant one.

On Artificial Analysis’s v5 Index, the top open-weights model, Kimi K3.1 Pro, scores 46 with its weights not yet released, and the best models with released weights score 32 against 58 for the leader. The framework was published in March 2026 and does not discuss this Index.

What Is Not Established

The scores are Artificial Analysis’s own measurements on its own suite, and its version history shows the evaluations inside the Index changed with v5. Each model is tested at named reasoning settings, so the same model can score and cost differently at lower settings.

The post does not say how the gap of 3 points under the previous version was measured. Meta’s Muse Wonder Full (max) scores 50 on the leaderboard, but no cost per task is listed for it. We did not contact Artificial Analysis or the model makers.

The Bottom Line

On Artificial Analysis’s Intelligence Index v5, Claude Opus 5.5 (max) leads at 58, one point ahead of GPT-6 Astra (max), and 60% of the Index now rests on private or held-out test sets. GPT-6.1 Sol (max) scores 55 at $1.27 per task against $10.04 for Opus 5.5 (max).