Is it worth the tokens? EstimatePrices: 11 Oct 2026

What one attempt is worth (reward after fees × your chance of winning − the attempt's token cost), and what that means for your day: how many attempts a daily cap buys, whether the cap or the supply of tasks is the limit, how long until the first payout and what an hour of your review time is worth. Presets come from economics.json (refreshed 12 October 2026, at list prices on 11 October 2026; list prices, not costs). Win rates, rewards and supply are measured; token costs are estimated; daily caps and owner minutes are assumed. The pool made no paid attempt and no model call against any task.

List prices on 11 October 2026; presets refreshed 12 October 2026. These are list prices, not costs. The TaskMarket presets use the 12 October reading (win chance 0.0155, 90% range 0.0117 to 0.0203); the MoltJobs preset is new (operator caveat: the founder account posted every funded job).

The reward

Withdrawal or gas cost per win; 0 where none is documented.

Your chance of winning

Which win rate to use

90% range of the win rate estimatedUsed for the EV range below. Wider of a Wilson interval and a bootstrap over tasks (Algora: bounds).

The cost of one attempt

Price band (per million tokens, list prices on 11 Oct 2026)

Paid routes, x402 charges, gas. 0 in every preset.

Your day: daily cap and owner time

What you let the agent spend on tokens per day.
Delays and per-payout time

Your day at this cap and price band

back per $1 of tokens at this band

    Per attempt: EV range at this price band

    Central EV at each price band

    Correction · 5 Oct 2026Run 2's presets (1 October) used settled tasks only, so they counted survivors. The old "TaskMarket top task" preset (15 USDC, 231 submissions) implied $0.060 per attempt; pooled over all closed TaskMarket tasks of 3 USDC or more it is $0.0031 (1 award in 1,484 submissions). Algora's median fell from $33.33 to $9.09 and Superteam's from $4.27 to $2.37 for the same reason.
    Every preset change since run 2
    PresetRun 2 (1 Oct)Run 6 (5 Oct)
    TaskMarket median: win chance1/87 = 0.0115, equal chance on 73 settled tasks0.0130 pooled on 81 closed tasks, 16 unawarded included
    TaskMarket median: expected payout$0.0213$0.024
    TaskMarket median: token costtyped-in guess, $0.01 to $3$0.32 code band / $0.072 single-file band at mid prices (estimated)
    TaskMarket p90 reward$2$4
    TaskMarket top task / larger tasks$0.060$0.0031
    Algora median$33.33$9.09 (lower bound)
    Algora crowded bounty$0.71$0.148 (estimated)
    Superteam agent-allowed$4.27$2.37
    Superteam recent$3.97$2.13 (n=9)
    Execution MarketEV $0.0174 at a win chance of about 1assignment: no win rate; margin −$0.080 per assigned job at mid prices
    Preset changes on 12 October 2026
    Preset11 Oct12 Oct (list prices on 11 Oct)
    TaskMarket typical: win chance0.01410.0155 (4 still-payable tasks settled, plus window roll-off)
    TaskMarket typical: central EV at mid band−$0.283−$0.280
    TaskMarket typical: tasks a day2.732.77
    TaskMarket single-file: central EV−$0.044−$0.041
    MoltJobs assigned job (new)not a preset+$0.45 margin per assigned job at mid band; supply 0 (no new job since 23 Sept)

    No economics verdict changed. Price tiers unchanged.

    Preset changes on 11 October 2026
    PresetBefore (7 Oct presets, 5 Oct prices)11 Oct (list prices on 11 Oct)
    Mid band output price$7.25 per million$6.75 per million (−6.9%)
    TaskMarket typical: win chance0.01360.0141 (window roll-off)
    TaskMarket typical: central EV at mid band−$0.291−$0.283
    TaskMarket typical: tasks a day3.132.73
    TaskMarket single-file: central EV−$0.047−$0.044
    Superteam agent-allowed: central EV$2.64$2.65
    Algora paying boards: central EV$8.77$8.78
    Execution Market: margin per assigned job−$0.080−$0.079
    DeskCrew $1 bounty (new)not a preset+$0.045 per entry at mid band, 0.06 entry fee included

    No economics verdict changed. Mid-tier returns rose 1–3%; budget-tier returns did not change.

    Not a prediction of income. Win rates are pooled over everyone who entered, including spam, and are snapshots. Costs use list prices without caching or batch discounts. The calculator ignores taxes, the time you spend checking work, and ban risk: never submit where a venue's rules forbid automated entries. A flat-rate subscription is quota-limited, not free compute and not resellable; check your provider's rules on automated use before relying on it.