OpenAI launches GPT 5.5 focusing on research and programming.

In its week of new features, the OpenAI introduced a new language model about a month after its launch. GPT 5.4. The new GPT 5.5 It's designed to be smarter and can even operate autonomously on computers. Check out the new features:

Greater efficiency compared to GPT 5.4

One of the major differences between language models is that the GPT 5.5 It can deliver better efficiency, especially in complex tasks, but in the same timeframe as the model launched in March of this year. The idea is that it can use fewer tokens with greater accuracy, making better decisions in the same timeframe that the previous language model required.

In practice, this provides a more complete experience without the need to wait longer for a complete response, even with more complex files. OpenAI announced that in Codex mode, a model focused on programming and software engineering, the GPT 5.5 revised 24.771 K-1 tax forms (from the US Internal Revenue Service), totaling 71.637 pages, using a workflow that excluded personal information, reducing human processing time by two weeks.

Autonomous computer use

The big difference is that the new model no longer needs step-by-step instructions for complex tasks. It can handle... The ambiguity of the real world, verifying one's own work, and making autonomous decisions. about which software tool to use next to accomplish a goal.

In practice, the system gained the ability to operate the computer independentlyArtificial intelligence can read information on the screen, move the cursor, and click buttons, correcting its own course if it encounters interruptions such as unexpected windows or layout changes. Simply activate the language model in... OpenAI Codex and let him do all the work while the developers handle other tasks.

This leap in autonomy has an immediate impact on software development and systems engineering. Senior testers report that the model demonstrates a true... "conceptual clarity", being able to understand the architecture of entire systems to identify flaws and suggest complete refactorings in a single pass.

The numbers reflect this evolution: the GPT-5.5 reached 82,7% accuracy in complex command-line flows (Terminal Bench 2.0) to 58,6% effectiveness in solving real-world GitHub problems (SWE-Bench Pro).

Support in complex research

In the academic field and in complex research, the GPT-5.5 assumes the role of a true "scientist". A OpenAI He highlighted that artificial intelligence has achieved cutting-edge results in bioinformatics benchmarks and in the analysis of genetic data.

The major practical milestone, however, occurred during internal company testing: a version of the model helped researchers discover a new mathematical proof about "Ramsey Numbers", solving an old and highly complex challenge in the field of combinatorics.

Despite this significant increase in reasoning ability, the model did not sacrifice its performance. Scientists who have already tested the new language model for internal OpenAI testing confirm that the language model can be considered a complex assistant that does more than just read content, but also provides practical solutions for everyday activities.

Early users tested GPT-5.5 Pro on ChatGPT less as a one-time response mechanism and more as a research partner: analyzing manuscripts in multiple stages, testing technical arguments, proposing analyses, and working with code, annotations, and the context of PDFs. The common thread is that GPT-5.5 is more effective at helping researchers move from question to experiment and, ultimately, to producing results.

OpenAI on the new GPT 5.5 language model.

Price and availability

For the end user, the GPT-5.5 It is now available immediately on ChatGPT for subscribers of the Plus, Pro, Business, and Enterprise plans. GPT-5.5 ProDesigned for demands requiring extreme precision, this feature is restricted to the highest-tier plans: Pro, Business, and Enterprise.

No CodexOn OpenAI's programming-focused platform, the release is even broader (including Edu and Go accounts) and features a robust context window of 400 tokens. One detail that stands out for development teams is the introduction of "Fast Mode," which generates responses 1,5 times faster but charges a toll of 2,5 times the standard value for this urgency.

For those who will integrate the technology into their own systems via API (Releasing soon), the official prices are:

• There are cost-effective tactical alternatives: modalities such as Batch e Flex They cut the bill in half, while priority processing requirements multiply the value by 2,5.
• GPT-5.5 (1 million token window): US$5 for every 1 million tokens entered and US$30 for every 1 million tokens exited.
• GPT-5.5 Pro: US$30 (entry) and US$180 (exit) for every 1 million tokens.

What did you think of the new features of the ChatGPT language model? Do you think it would help you in your daily life? Tell us in the comments! Comment!

See also other features

With information: OpenAI

reviewed by Luis Antonio Costa in 23 / 04 / 26