OpenAI has introduced a new model called GPT-6.1 Terra, which the company says is designed to lower the cost of running artificial intelligence. According to Techworld Bangladesh, OpenAI claims the new version can complete certain tasks at half the expense of its predecessor.
The company also says the model has improved at reasoning, interpreting images and carrying out tasks on a computer automatically. The release comes as competition among developers of large language models intensifies.
Why it matters
Cost has become a central battleground in the AI industry. Running large models at scale is expensive, and businesses weighing whether to deploy AI in their operations look closely at price per task as well as raw capability.
Techworld Bangladesh reports that OpenAI and Anthropic have both cut prices for their large language models by as much as 80 percent as competition in the open-weight segment has heated up. A model that promises the same work for less money fits directly into that trend.
The launch also matters because it pairs cost savings with capability gains, at least on OpenAI's own measures. If the claims hold up, organisations that previously found AI too costly for high-volume tasks may have a new option to consider.
Benchmark gains in image analysis
OpenAI says GPT-6.1 Terra has improved its ability to analyse images. On the MMMMU-Pro benchmark, which tests multimodal understanding, the new model scored 85.5 percent, according to Techworld Bangladesh. The earlier model scored 81.7 percent on the same test.
The company further claims the new model is less prone to producing incorrect information. At its highest reasoning setting, the error rate stands at 4.3 percent, per OpenAI's figures.
On safety, OpenAI says it has added the security system used in GPT-6 Astra to the new model, part of its effort to address cybersecurity risks.
A weaker showing on computer-use tasks
The picture is more mixed on tasks that involve operating a computer. Techworld Bangladesh reports that GPT-6.1 Terra scored 74.2 percent in that category of testing. Anthropic's Claude Sonnet 5.5 scored 83.9 percent on the same measure.
That gap suggests OpenAI's new model does not lead across every benchmark, even as the company emphasises its efficiency gains. The comparison also underlines how closely the two companies are being measured against each other.
An analyst's caution on true cost
Not everyone is convinced that a lower token price tells the full story. The chief analyst at Greythound Research told Techworld Bangladesh that token pricing alone cannot determine the real cost of using AI.
The analyst argued that other expenses must also be counted: checking results, correcting mistakes and integrating AI with systems that are already in place. Those costs can add up quickly in real deployments, particularly where accuracy matters.
The point is a reminder that headline price cuts and benchmark scores describe only part of what a business spends when it puts a model to work.
What to watch
Whether independent testing confirms OpenAI's cost and accuracy claims for GPT-6.1 Terra, and whether Anthropic responds with pricing or performance changes of its own.
