AI Agent StoreHigh
Tested setting0.0 pts vs High
OpenAI · REASONING EFFORT GUIDEGPT-5.4
GPT-5.4
How much thinking?
Compare None vs Low vs Medium vs High vs Extra high where tested. Find the effort setting that fits your task, and see how much extra reasoning changes the result.
SAME MODEL. DIFFERENT THINKING.
Choose a thinking level for GPT-5.4
LiveBench
AI Agent Store task-weighted category mix. An editorial selection aid, not a measured agent success rate; weights match our published Agent Fit methodology.
Explicit effort rows · Question set: 2026-01-08 · Reviewed 2026-10-15Check the source ↗SUGGESTED SETTING TO TESTExtra high
Lowest tested effort within 1.0 points of the best tested general agent fit.
77.6 / 100Exact best1.0 pts10 pts
Set 0 for the highest score. Increase the gap to explore cheaper or lighter options.
THE EFFORT CURVE
Does more thinking help?
● Near-best pickOrange = score decreased
Scores out of 100 · Full 0–100 scale · tap a point to select mode B. Cyan band: within 1.0 points of the best tested score.
This study has no comparable cost data.
The recommendation uses the lowest tested effort within your allowed score gap. That does not prove it is the cheapest or fastest.
Use the token calculator below with your own usage measurements. Effort labels alone do not tell us a token bill.
Estimate your own token costs ↓Score difference · B minus A+7.0 pts
Large gap · B scored higher
Highest minus lowest score in this study7.0 pts
Best minus lowest tested score
Highest observed score77.6
Extra high
We describe gaps under 1 point as small, 1–under 5 as noticeable, and 5 or more as large. These are reading aids, not proof that a difference would hold up in repeated tests. Equal scores can still hide different mistakes.
EVERY TESTED MODE
2 modes · untested settings are left blankResults for each tested setting
70.6 / 100
Extra high
Suggested setting+7.0 pts vs High
Full benchmark table & exact configuration IDs
Swipe inside the table for all measures. Figures from different study selections are not directly comparable.
| Effort | Overall benchmark | General agent fit | Coding agent fit | Research agent fit | Writing & support fit | Reasoning | Coding | Agentic Coding | Mathematics | Data Analysis | Language | Instruction following | Cost / task | Median seconds | Output tokens | Exact source ID |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| High | 75.1 | 70.6 | 63.8 | 78.5 | 75.3 | 85.7 | 78.2 | 46.7 | 90.0 | 77.0 | 83.0 | 65.0 | — | — | — | gpt-5.4-high |
| Extra high | 80.3 | 77.6 | 75.0 | 81.1 | 77.9 | 88.1 | 77.5 | 70.0 | 94.1 | 79.3 | 82.6 | 70.2 | — | — | — | gpt-5.4-xhigh |
SCALE THE OBSERVED COST
Effort cost difference calculator
What would A versus B cost if you repeated this benchmark’s average workload?
A · High—
B · Extra high—
Difference—This study did not report comparable costs
This source has no comparable dollar-cost observation. Use the token calculator below with your own usage; token counts alone cannot supply a full bill.
YOUR WORKLOAD
Reasoning-token cost calculator
Estimate the extra bill from thinking tokens. The starting token counts are editable examples, not measurements of GPT-5.4. Rates start from the current model snapshot.
A · High$55estimated monthly cost
B · Extra high$115estimated monthly cost
Mode B costs $60 more per month with these assumptions.
This estimate uses the token counts and prices you enter. Excludes retries, tools, caching, long-context premiums, hosting and taxes. Includes thinking in output billing; do not also include it in “visible output.” Check endpoint limits separately. Changing mode does not automatically predict tokens or quality.
GPT-5.4 effort benchmarks at a glance
Each card compares settings tested together in one study. Scores from different benchmarks or question sets measure different work and are not averaged together.
General tasks · LiveBench 2026-01-08 · Explicit effort rows
AI Agent Store task-weighted category mix. An editorial selection aid, not a measured agent success rate; weights match our published Agent Fit methodology.
- High
- 70.6
- Extra high · pick
- 77.6
Near-best pick: Extra high within 1 point of the best score. Highest-to-lowest score gap: 7.0 points.
Explore this study ↗All results and the exact settings tested
A dash means the study did not report that result. This table compares only the versions and settings tested on the same questions in this study.
| Effort / budget | Overall benchmark | General agent fit | Coding agent fit | Research agent fit | Writing & support fit | Reasoning | Coding | Agentic Coding | Mathematics | Data Analysis | Language | Instruction following | Exact source configuration |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| High | 75.1 | 70.6 | 63.8 | 78.5 | 75.3 | 85.7 | 78.2 | 46.7 | 90.0 | 77.0 | 83.0 | 65.0 | gpt-5.4-high |
| Extra high | 80.3 | 77.6 | 75.0 | 81.1 | 77.9 | 88.1 | 77.5 | 70.0 | 94.1 | 79.3 | 82.6 | 70.2 | gpt-5.4-xhigh |
Math, code & science · OckBench
Correct answers out of 200 problems. Domain proportions are 100 math, 60 coding and 40 science.
- None
- 23.5
- Low
- 78.0
- Medium · pick
- 82.0
- High
- 80.5
Near-best pick: Medium within 1 point of the best score. Highest-to-lowest score gap: 58.5 points.
Explore this study ↗All results and the exact settings tested
A dash means the study did not report that result. This table compares only the versions and settings tested on the same questions in this study.
| Effort / budget | Overall accuracy | Mathematics | Coding tests | Science | Average output tokens | Exact source configuration |
|---|---|---|---|---|---|---|
| None | 23.5 | 17.0 | 20.0 | 45.0 | 689 | gpt-5.4 |
| Low | 78.0 | 73.0 | 91.7 | 70.0 | 3,056 | gpt-5.4-low |
| Medium | 82.0 | 78.0 | 93.3 | 75.0 | 3,177 | gpt-5.4-medium |
| High | 80.5 | 74.0 | 98.3 | 70.0 | 3,267 | gpt-5.4-high |
Reasoning effort, explained
Which GPT-5.4 thinking level is most efficient?
Extra high is the lowest-effort option within 1 point of the best general agent fit in LiveBench. The tested score range is 7.0 points. This allows a score up to 1 point below the best result in the first listed study. Change the study, task and allowed score gap above to explore another comparison. It is a shortlist for testing your own workload.
Which GPT-5.4 effort settings are compared?
None, Low, Medium, High, Extra high appear in the collected studies. Each chart only includes modes tested together in that study. These published labels do not establish that every endpoint or application exposes every setting.
Does more effort improve GPT-5.4?
In the first listed LiveBench sweep, the tested score range is 7.0 points and no adjacent effort increase lowers the score. This does not guarantee a gain on every task or with a different testing setup.
How much more does high effort cost for GPT-5.4?
Use the A/B selector and measured-cost calculator when the source reports dollar costs. For your own agent, enter average input, visible output and thinking tokens in the token calculator. Higher effort is adaptive, so the label itself cannot tell you the bill. Missing costs stay missing.