GPT-6.1 Sol vs Terra: Sol costs 2x and scores 4 points higher

GPT-6.1 Sol vs Terra comes down to price and depth. Both are OpenAI models with the same 1,050,000-token context window, but Sol costs twice as much per token and scored 52 on Artificial Analysis's Intelligence Index, against 48 for Terra. Terra is the faster one. OpenAI released it on October 26, 2026, a month after Sol, which came out on September 29.

OpenAI's changelog describes Sol as the model for complex coding and professional work. It describes Terra as a fast, efficient model for visual reasoning, precise computer use and clear, natural writing. Our own report on the Terra launch has the details of that release.

This comparison is based on the models' docs and published results. We did not run either model ourselves.

GPT-6.1 Sol vs Terra at a glance

Prices are per million tokens, as of October 26, 2026, from OpenAI's model pages for Terra and Sol. Independent numbers are from Artificial Analysis, read on October 26, and are for each model at its "max" reasoning setting.

How much GPT-6.1 Sol and Terra cost

Terra's list prices are exactly half of Sol's on every line: $1 against $2 for input, $5 against $10 for output. Both apply a surcharge to long prompts. Above 272,000 input tokens, input and cache rates double and output costs 1.5 times as much for the whole request. Batch and Flex processing cost 50% less than Standard on both, Fast mode costs twice Standard, and regional processing adds 10% where available.

Only Sol has an "ultrafast" mode, at six times Standard. OpenAI added it to the Responses API on October 8 with service_tier: "ultrafast", to cut the time between output tokens.

The price gap is smaller in practice than the list prices suggest. Artificial Analysis puts the cost of running its full index at about $0.72 for Sol and $0.40 for Terra, a gap of about 1.8 times. Its blended price, which mixes input and output tokens, is $1.47 per million for Sol and $0.73 for Terra.

On OpenRouter, the Sol page we could read lists one route: Azure in Europe at $2.20 input and $11 output per million tokens, available only on Business or Enterprise plans. The Terra page showed no pricing, and EU in-region routing is not available for it. We could not read OpenRouter's standard global prices for either model.

Which is smarter and which is faster

On the Artificial Analysis Intelligence Index, Sol scored 52 and Terra 48. That put Sol 13th of 221 models and Terra 25th. The median for comparable models on the page was 28. Four points is a real gap but a small one next to the distance both sit above the median.

Terra is faster: 90.9 output tokens per second against 69.2 for Sol. Terra is above the 83.4 median on the page, and Sol is below it.

These are scores for the "max" reasoning setting, the most expensive one. At lower settings, both the scores and the costs will differ, and Artificial Analysis's page did not show them.

OpenAI's own claim for Terra is a drop in the error rate on prompts built to induce mistakes, from 53.6% for GPT-5.6 Terra to 16.7% at low reasoning effort, according to the launch coverage we read. That compares Terra with its predecessor, not with Sol, and it is OpenAI's own test. We could not open the announcement page itself, which returned an access error, so we have not checked this against the original.

Features and limits that match, and ones that don't

The two models are close on features. Both take text and images and return text only. Both support streaming, structured outputs, function calling, file search, web search and prompt caching. Neither supports fine-tuning or predicted outputs. Both work through the Responses and Chat Completions APIs, and OpenAI says tool calling should go through Responses. Standard rate limits are the same on the pages: 5,000 requests and 1 million tokens per minute on the Build tier, 10,000 and 4 million on Launch, 15,000 and 40 million on Grow.

The differences are small. Terra has a later knowledge cutoff, June 1 against April 30. Terra accepts a reasoning effort of "none", which Sol's page does not list. Sol's changelog entry mentions a beta multi-agent feature that lets the model hand work to subagents in a Responses API request. Terra's does not.

Which one to pick

Pick Sol for the hard jobs: large coding changes and professional work where an extra four index points matter more than cost, and where OpenAI aims the model. Pick Terra for volume, for latency-sensitive work, and for tasks involving images, computer use and writing, where OpenAI says it was built to be strong and where half the price adds up.

A sensible test is to run your own prompts on both at the same reasoning setting and compare cost per correct answer. A general index does not know your workload. Artificial Analysis and OpenRouter will fill in more numbers as both models get used, so check them again before committing.