GPT-6.1 Sol vs Claude Opus 5.5: cost per task against top scores

The short answer: Sol for cost per task, Opus for the top index score

Choose GPT-6.1 Sol if you run many hard tasks and cost per finished task matters: Artificial Analysis measured it within 4 points of Opus 5.5 at about an eighth of the cost per task. Choose Claude Opus 5.5 if you need the highest overall score in that index or want it in a chat app on a $20 Claude plan. Sol lives in ChatGPT Work and Codex rather than regular chats.

Scope and evidence

Task: complex coding, research and agent work through APIs or each vendor's apps. Prices are in US dollars from OpenAI's Sol page and help center and Anthropic's Opus 5.5 overview, Effort page and Claude pricing page; Artificial Analysis supplied the index scores and per-task costs, as published in its v5 announcement and read on October 28, 2026.

This page reports published figures only. Artificial Analysis measured both models at max effort; your results at other settings will differ.

Compare the same criteria

Each criterion contrasts OpenAI's mid-tier model with Anthropic's upper-tier one, then explains the practical effect.

• API price per million tokens. Sol: $2 input, $0.10 cached, $10 output. Opus 5.5: $4 input, $0.20 cache read, $20 output. What this means: per token, Opus is double the price.
• Intelligence Index v5 at max effort. Sol: 55. Opus 5.5: 58, the top score. What this means: a 3-point gap overall.
• Cost per index task at max effort. Sol: $1.27. Opus 5.5: $10.04. What this means: by our reading, Opus also used more tokens per task, since the gap is wider than list prices.
• Selected evaluations. Terminal-Bench-Science: Sol 58%, Opus 59%. CritPt v2 hard subset: Sol 69%, Opus 61%. HLE-Diamond: Opus leads at 68%. What this means: neither leads everywhere.
• Reasoning control. Sol: effort low to max, default medium. Opus 5.5: adaptive thinking always on, effort low to max, default medium. What this means: equivalent ranges.
• Context and output. Sol: 1.05M tokens, 128K output. Opus 5.5: 1M tokens, 128K output. What this means: practically equal.
• App access. Sol: ChatGPT Work and Codex on paid plans, 15 to 160 local messages per five hours estimated for Plus. Opus 5.5: Claude Pro and Max, not Free. What this means: Opus is available as a chat model.

View image

Which fits your real work?

Nightly agent runs over hundreds of repositories. Sol fits, since cost per task drives the bill and the index gap is small.

A physics or maths research workflow with code execution. Sol fits on the CritPt v2 result, 69% against 61% for Opus.

A one-off high-stakes analysis where every point of accuracy counts. Opus 5.5 fits as the top model in the index overall.

Cost of getting the work done

API: 2 million input and 200,000 output tokens cost $6 on Sol and $12 on Opus 5.5 before caching. In Artificial Analysis's runs at max effort the gap per completed task was about eightfold, $1.27 against $10.04.

Apps, per month: ChatGPT Plus at $20 includes Sol in Work and Codex; Opus 5.5 comes with Claude Pro, which Anthropic prices at $20 for a monthly subscription, with Max plans from $100.

Try both on the same small task

Suggested check: run one research question that needs a short script, since index results split on science tasks.

• The script runs and reproduces the stated number.
• Every assumption from the prompt is used, and any extra one is named.
• Units are consistent throughout.
• Compare tokens used and run each model twice before concluding.

Decision checklist

Use the conditions that apply to you. Profiles: our GPT-6.1 Sol profile and our Claude Opus 5.5 profile.

• If cost per completed task drives the decision, start with Sol.
• If you need the top overall score in Intelligence Index v5, start with Opus 5.5.
• If you want the model in a chat window, Opus 5.5 is on Claude Pro.
• If you already code in Codex, Sol is included in your ChatGPT plan.
• If your work is maths or physics with code, compare both on CritPt-style tasks.

Go a little further

Related reading

• GPT-6.1 Sol profile →
• Claude Opus 5.5 profile →