prediction-fees

Exact trading fees for Kalshi and Polymarket, from a dated schedule. No dependencies, MIT.

pip install prediction-fees

from prediction_fees import kalshi_fee, kalshi_fee_levels, polymarket_fee, polymarket_rebate

kalshi_fee(0.50, 100)  # Decimal('1.75')  taker
kalshi_fee(0.50, 1)  # Decimal('0.02')  0.0175 rounds up to the cent
kalshi_fee(0.50, 100, maker=True)  # Decimal('0.44')
kalshi_fee(0.50, 100, as_of="2026-11-03")

# one order that walked three levels: summed first, rounded once
kalshi_fee_levels([(0.45, 10), (0.46, 20), (0.48, 5)])  # Decimal('0.61')

# a series without maker fees (Kalshi GET /series/{ticker} fee_type "quadratic")
kalshi_fee(0.50, 100, maker=True, fee_type="quadratic")  # Decimal('0.00')

polymarket_fee(0.50, 100, category="crypto")  # Decimal('1.75000') USDC
polymarket_fee(0.50, 100, rate=0.03)  # Decimal('0.75000')  the market's own feeSchedule.rate
polymarket_fee(0.50, 100, maker=True)  # Decimal('0.00000')
polymarket_rebate(0.60, 100, category="politics")  # Decimal('0.240000')  maker rebate

Every function returns a Decimal. Prices are in dollars (0 to 1). Inputs can be int, float, str or Decimal; floats are read through their shortest repr, so 0.07 is exactly 0.07 and there is no float error in the round-up (a naive math.ceil charges a cent too many on, for example, 100 contracts at 0.10).

The rules

Kalshi. Per order execution (one order matched at one time, across however many price levels it took):

fee = round_up_to_cent( rate × multiplier × Σ C × P × (1 − P) )

rate is 0.07 for takers and 0.0175 for makers. multiplier is the series' fee_multiplier and applies to takers and makers (1 for most series; some 0.5, a few 0). Makers pay only on series whose fee_type charges makers (quadratic_with_maker_fees, quadratic_with_combo_maker_fees); a quadratic series charges makers 0. On 2 Nov 2026 that was 163 of Kalshi's 15,600 series. Read both from Kalshi's public GET /trade-api/v2/series/{series_ticker} (the series is the market ticker up to the first -) and pass them: fee_type="quadratic" makes a maker fill cost 0, and kalshi_multipliers(fee_type, fee_multiplier) gives the (taker, maker) multipliers. Without fee_type makers are charged (the conservative side).

Polymarket. Takers, per execution:

fee = round_down_to_0.00001( rate × Σ C × P × (1 − P) )

rate is the market's own coefficient: Gamma GET /markets?id= returns it as feeSchedule.rate (0 when feesEnabled is false), and some older sports markets still carry 0.03. Without it, the category table:

Pass the market's own coefficient as rate or category (case-insensitive); polymarket_market_terms(market) reads a Gamma row (its feeSchedule, else the category its feeType names, e.g. politics_fees). Fees are rounded to 5 decimals, down: Polymarket's docs say anything below 0.00001 USDC rounds to zero.

Makers pay 0 and earn a rebate, polymarket_rebate:

rebate = rebate_share × rate × Σ C × P × (1 − P)

Each market's daily rebate pool is its rebate share of the taker fees paid there, split pro rata by each maker's fee-equivalent, so a maker's rebate is exactly the share of its own fee-equivalent. Pass the market's feeSchedule.rebateRate as rebate_share (or a category). The rebate is not rounded (it is paid daily in aggregate) and the $1 minimum payout is not modelled.

Edge cases. A price of exactly 0 or 1 costs nothing (P × (1 − P) is 0). A price outside [0, 1], a negative size or rate, NaN or infinity raises FeeError (a ValueError). Zero contracts cost nothing.

Dates

The table (fees.json, shipped inside the package) lists each venue's rates with the date they took effect. as_of= takes a date, a datetime (read in UTC) or an ISO string and picks the entry in force that day; without it you get the latest entry. A date before the first entry raises FeeError rather than guessing; pass rate= to price an older trade.

prediction_fees.table() returns the whole table, sources and notes included.

What is not certain

Checked on 2026-11-01 (Polymarket and the Kalshi series rule on 2026-11-02):

• Kalshi's taker formula and 0.07 are well established. The maker coefficient 0.0175 and the 2026-07-07 effective date come from secondary sources quoting Kalshi's fee schedule PDF (kalshi.com/docs/kalshi-fee-schedule.pdf), which could not be fetched to confirm.
• Kalshi's API docs describe fees accrued to $0.000001 with a per-order rounding accumulator that refunds over-rounding. This library follows the published schedule (round up to the cent per execution), so a fill split across several executions may show a cent or two more than Kalshi's final charge.
• Polymarket's coefficients and rebate shares are as read from docs.polymarket.com on 2026-11-02. 2026-07-10 is the date of the last change in the published timeline (sports 0.03 to 0.05); earlier periods are not tabulated, and a market's own feeSchedule can still differ from the table (0.03 on some older sports markets), so prefer it. The docs say fees are rounded to 5 decimals and that anything below 0.00001 USDC rounds to zero; this library reads that as rounding down, which the docs do not state in so many words.
• Not covered: the $1 minimum rebate payout, taker volume rebates, Polymarket US (a separate exchange with its own schedule), Kalshi incentive programs and perpetuals.

Corrections with a source are welcome: hello@langbot.ai.

The same table and functions ship for JavaScript/TypeScript as prediction-fees on npm.

License

MIT. Maintained by Langbot.