OpenAI has rolled out GPT‑5.4 to ChatGPT, positioning the new model as its most capable and efficient system for professional work to date. The release lands just days after GPT‑5.3 Instant, signaling a rapid cadence aimed at both everyday users and high‑demand enterprise teams.
Alongside the base model, OpenAI introduced GPT‑5.4 Pro for peak performance and GPT‑5.4 Thinking for more controllable, stepwise problem‑solving. The company says power users should notice improvements immediately in ChatGPT and via the API, with paid access required at launch.
Here are five upgrades that matter right now—and what they unlock for real workflows.
Faster Results Through Improved Token Efficiency
OpenAI describes GPT‑5.4 as its most token‑efficient reasoning model yet, using significantly fewer tokens to reach the same or better outcomes compared to GPT‑5.2. In practice, that means lower latency and less “thinking out loud” to land on a high‑quality answer—useful for users on usage‑based billing where token count drives cost.
Consider a remediation plan from a 50‑page security audit: previous models often needed verbose intermediate steps to stay organized. GPT‑5.4 trims that overhead, producing structured, numbered actions more directly while preserving traceability. The efficiency also benefits long‑context tasks like RFP synthesis and literature reviews.
Stronger Performance On Complex Knowledge Work
On GDPval—a benchmark that scores well‑specified knowledge work across 44 occupations—OpenAI reports GPT‑5.4 sets a new bar, matching or exceeding industry professionals in 83.0% of comparisons. By contrast, GPT‑5.2 reached 70.9%. While benchmark design always matters, the gap suggests a measurable lift in reliability for tasks like drafting policies, writing product specs, or assembling market analyses.
Teams see the upside when instructions are concrete: “Redline this clause for indemnification risk,” or “Produce a 1‑page executive brief with a risk/impact matrix.” GPT‑5.4 appears better at honoring constraints and formatting, reducing the polish required to reach publication‑ready output.
Fewer Hallucinations And Cleaner Claims Overall
OpenAI says GPT‑5.4 is 33% less likely to make false claims and its claims are 18% less likely to contain errors versus GPT‑5.2, based on internal evaluations. That does not eliminate hallucinations, but it narrows the risk envelope in high‑stakes settings.
For example, when summarizing earnings calls or legislative drafts, a lower propensity for invented figures and misattributions reduces rework and the need for extensive fact triage. The takeaway for ops leaders: keep automated checks and human review in place, but expect fewer show‑stopping corrections.
New Thinking Mode For Midstream Steering
GPT‑5.4 Thinking can present an upfront plan, then continue working while you steer it mid‑response. Instead of waiting for a full answer before course‑correcting, users can nudge the approach in real time—“prioritize cost over speed,” “use a decision tree, not a flowchart,” or “target SQL joins, not spreadsheet merges.”
This guardrailed transparency helps align outcomes without added turns, especially in multi‑step tasks like data transformations, compliance checklists, or multi‑criteria product roadmaps. It’s effectively a collaboration loop that compresses the edit cycle and conserves tokens.
Deeper Office Integration And The Pro Tier
Enterprise users gain a new ChatGPT for Excel add‑in, allowing GPT‑5.4 to create, audit, and analyze spreadsheets from within familiar tooling. That includes generating formulas, building pivot tables, cleaning messy CSVs, and drafting scenario models tied to clear assumptions—a boon for finance, ops, and sales teams.
For teams seeking maximum headroom, GPT‑5.4 Pro targets the highest performance envelope. The new models are rolling out to ChatGPT and the API, with access gated to paid subscriptions initially. Combined with existing enterprise controls, the release underscores OpenAI’s push to make large‑scale deployment both feasible and governable.
Bottom line: GPT‑5.4’s speed, accuracy gains, controllable reasoning, and tighter spreadsheet workflows point squarely at knowledge work at scale. With competitors also advancing rapidly in enterprise AI, these upgrades are less about novelty and more about shaving minutes and mistakes from thousands of daily tasks—the kind of compounding effect that moves productivity metrics in the right direction.
