What Is GPT-5.4 Thinking?

OpenAI has released its latest frontier reasoning model, GPT-5.4 Thinking, alongside a detailed system card documenting the model's capabilities, safety evaluations, and limitations. The release marks another step in OpenAI's push to develop AI systems capable of tackling complex, multi-step problems through extended reasoning chains before delivering final answers to users.

Unlike standard language models that generate responses token-by-token without deliberation, GPT-5.4 Thinking uses chain-of-thought reasoning — working through problems internally before committing to an output. This architecture enables the model to handle mathematical proofs, complex coding tasks, scientific reasoning, and nuanced logical analysis with substantially greater accuracy than earlier systems.

The system card, which OpenAI publishes for all frontier models, provides a transparent view of how the company evaluates AI before deployment. It covers safety benchmarks, red-team results, potential misuse risks, and the specific mitigations implemented — giving researchers and enterprise customers the information they need to assess appropriate use cases for the new model.

Safety Evaluations and Red-Teaming Results

Safety testing for GPT-5.4 Thinking followed OpenAI's Preparedness Framework, evaluating the model across cybersecurity threats, biological and chemical weapons enablement, radiological risk, and autonomous resource acquisition. The system card places GPT-5.4 Thinking in the Medium overall risk category, meaning it can be deployed with standard safety mitigations in place without triggering additional restrictions.

Red-team evaluations tested the model's resistance to jailbreaks, indirect prompt injection, and multi-step adversarial manipulation. GPT-5.4 Thinking demonstrated improved resistance to many attack vectors compared to prior generations, though it remains imperfect against highly sophisticated adversarial inputs — a caveat that applies to all current AI systems regardless of training sophistication.

Evaluations of persuasion and manipulation capabilities found that the model's safety training substantially reduces its willingness to produce content designed to deceive or coerce users. OpenAI also evaluated behavior in agentic settings, where the model might take sequences of actions with real-world consequences, and found performance within acceptable safety parameters for the Medium classification threshold.

Benchmark Performance and Capabilities

On standard reasoning benchmarks, GPT-5.4 Thinking shows meaningful improvements over its predecessor. The model achieves state-of-the-art results on MATH and competitive programming evaluations, and demonstrates strong performance on scientific reasoning tasks that require integrating information across multiple domains. Graduate-level academic questions in physics, chemistry, and formal logic show particular strength relative to prior-generation models.

The extended thinking window — the amount of internal computation the model performs before outputting a response — has been increased compared to earlier versions. This allows GPT-5.4 Thinking to tackle problems requiring sustained multi-step analysis rather than single-hop inference. For enterprise deployments, this translates into more reliable performance on complex workflows like financial modeling, code review, and research synthesis tasks.

Despite these improvements, the system card is explicit that GPT-5.4 Thinking is not infallible. The model can still hallucinate facts, make arithmetic errors on sufficiently complex calculations, and produce overconfident answers where its training data is sparse or ambiguous. OpenAI recommends human oversight for high-stakes applications and cautions against using the model as a sole decision-maker in critical systems.

Chain-of-Thought Transparency

One of the more technically significant aspects of the system card is its treatment of chain-of-thought transparency. OpenAI continues its policy of showing users portions of the model's reasoning process, allowing verification of the logic path taken to reach a conclusion. This transparency serves a safety function by making hidden deceptive reasoning structurally harder, and a practical function by helping users identify where model logic diverged from their own expectations.

The system card acknowledges limitations in using visible chain-of-thought as a complete safety guarantee. Research published in parallel with this release found that what reasoning models display in their thinking traces does not always perfectly correspond to the underlying computational process. OpenAI is continuing to investigate whether visible reasoning accurately reflects true internal decision pathways — a question with deep implications for AI interpretability and oversight.

This transparency effort connects directly to broader safety research within OpenAI on whether reasoning models can be instructed to suppress or falsify their thinking. Evidence suggests this is structurally difficult for current architectures, a finding that reinforces the value of chain-of-thought monitoring as a real signal rather than cosmetic output theater.

What GPT-5.4 Thinking Means for Enterprise AI

For organizations deploying AI in complex workflows, GPT-5.4 Thinking represents a meaningful capability upgrade over previous reasoning models. Improved reasoning makes it better suited for tasks that currently require extensive human review — contract analysis, scientific literature synthesis, complex debugging, and multi-document summarization with nuanced synthesis requirements.

Enterprise API access is available through OpenAI's standard pricing tiers. Extended thinking is available at higher token costs reflecting the additional compute involved, a tradeoff that organizations will need to evaluate against the quality improvements for their specific use cases. OpenAI has committed to ongoing safety monitoring and will update the system card as new capabilities or risks are discovered through deployment.

The release continues a pattern of OpenAI publishing detailed safety documentation alongside capability releases — a practice that sets a transparency standard other major AI developers are under increasing pressure to match. As reasoning models become core infrastructure for enterprise AI, the quality and depth of these evaluations will become an important factor in procurement and deployment decisions across industries.

This article is based on reporting by OpenAI. Read the original article.