Betaflight Anti Gravity: what it does on throttle punches and how to tune it

Most pilots never touch it, which is fine, because it is on by default. The reason to look at it is that the default is strong, the scale changed in 4.4, and old tuning advice with numbers like 3.5 or 3500 no longer fits current firmware.

What the feature does

The Betaflight 4.4 pull request that rewrote Anti Gravity (PR #11679, by ctzsnooze, building on work by limonspb) describes the current behavior. Reading the 2026.6.2 source code fills in the details:

• The flight controller looks at how fast the throttle is changing and runs that signal through a PT2 low-pass filter.
• The signal is weighted towards the lower half of the throttle range, and chops count more than punches of the same size. The code multiplies the throttle derivative by the square of (1 - throttle) and adds an extra term only when the throttle goes up, which means moves near full throttle get less boost.
• The filtered value is added to the I gain on roll and pitch. Yaw gets no I boost since 4.4.
• A smaller P boost is applied on roll and pitch only. It fades when you are giving stick input: above 50 degrees per second of requested rotation it starts to shrink, and at 100 degrees per second it is half as strong. The idea is to help during straight throttle pumps without making flips and rolls feel different.
• The I term limits that stop windup also apply to the boost, so Anti Gravity cannot push the I term past them.

The 4.4 release removed the old Step mode, so there is only one way it works now. The 4.2 tuning notes also point out that the gain is independent of your I values. Changing I does not change how strong Anti Gravity is.

The settings

All three settings are stored per PID profile. The values below are from Betaflight 2026.6.2 and are the same as in 4.4 and 4.5:

Setting	Range	Default	What it does
anti_gravity_gain	0-250	80	Strength of the I boost. 0 turns it off.
anti_gravity_p_gain	0-250	100	Strength of the P boost. 100 is normal, 0 is none.
anti_gravity_cutoff_hz	2-50	5	Cutoff of the filter on the throttle signal.

Some numbers to put the default in context. According to the pull request, a gain of 80 means about eight times the default I gain during the fastest throttle change. The P boost is roughly a quarter of the I boost, and on defaults P briefly climbs to about 2-2.5 times its normal value.

The 4.4 release notes and the pull request text both give 6 Hz as the default cutoff. The firmware itself shipped with 5 Hz in 4.4.0 and still uses 5 Hz in 2026.6.2. Trust get anti_gravity_cutoff_hz over any guide.

Old values on the new scale

Before 4.4 the gain was stored in thousands. The default was 3500 in the CLI, shown as 3.5 in Configurator, and guides from that time quote numbers on that scale. Oscar Liang's older tuning guide, for example, says values between 4 and 6 worked for his quads, and the Freestyle Tuning Principles page on the Betaflight wiki still suggests 3.5-5 for a 5" quad. The pull request gives the conversion: a CLI gain of 35 on the new scale equals the old 3500. So never paste an old CLI value like 3500 into current firmware.

In Configurator, the PID Tuning tab has an Anti Gravity switch and a Gain field. On 4.4 and newer the field shows the CLI value divided by 10, so the default 80 appears as 8.0, and the help text explains it as "about 8x iTerm boost". Turning the switch off sets the gain to 0.

Do you need to change anything

On a typical freestyle quad, usually not. The pull request calls the default "quite strong" and warns against raising it without checking for short oscillations on full throttle punches and chops. It also says the default may be too much for highly tuned race builds, which are the quads most likely to need a lower value. Configurator's help text points the other way for some builds: a higher gain may help quads with low control authority or with an offset center of gravity, for example a heavy camera mounted far forward.

The pull request lists one known side effect. Because the boost reacts to throttle movement, a quad can feel a bit laggier when you first push the throttle up from zero.

A quick test in the field

The pull request suggests a simple check, with no logging needed:

1. Fly straight forward with a reasonable pitch angle.
2. Punch the throttle quickly, then chop it, with roll and pitch sticks centered.
3. Watch for the nose wandering up or down and for wobbles on pitch or roll.
4. Repeat with a higher and a lower anti_gravity_gain.
5. Compare against the gain set to 0, so you know what the feature is actually doing on your quad.

The PID Tuning Guide on the Betaflight wiki, a set of community notes, gives the same rule in one line: punch the throttle and cut it, and if there is movement, add Anti Gravity gain. Keep the pull request's warning in mind while doing that. If the slow drift turns into a short, fast wobble right after the punch or chop, you have gone too far.

A more careful method with Blackbox

If you log with Blackbox, the pull request gives a more formal order:

1. Set anti_gravity_p_gain = 0, so you tune the I part alone.
2. Set debug_mode = ANTI_GRAVITY and record a few punches and chops.
3. Try anti_gravity_gain in steps of 1, 40, 80 and 120. Pick the value that gives the best result without starting an I term wobble.
4. Bring the P boost back and try anti_gravity_p_gain at 50, 100 and 150.

With this debug mode, the first two fields show the throttle derivative before and after the filter. In 2026.6.2 the third field is the I multiplier and the fourth is the P multiplier, both times 1000, so 1000 means no boost and 2000 means double. The pull request text says times 100, but the current code uses 1000.

When to move the cutoff

The pull request calls the cutoff very sensitive. A higher cutoff makes the boost stronger but shorter, a lower one makes it longer and weaker. Its advice:

• Heavy, slow quads that wobble for longer than about 80 ms after a throttle change may do better with a lower cutoff.
• Racers may get the best result at 6 or 7 Hz.

Change one setting at a time, and keep the gain fixed while you test the cutoff.

Turning it off and seeing it work

The feature itself is switched with feature ANTI_GRAVITY, which is on by default. If you turn the feature off and at least one PID profile still has a gain above 0, the Modes tab shows an ANTI GRAVITY mode, so you can enable it with a switch. That is a convenient way to do the on-off comparison in a single flight.

The OSD has an "Anti gravity" element. It shows "AG" while the boost is above a small threshold, which tells you whether a wobble you see in the goggles lines up with the boost.