Works off each platform's own published per-model pricing and credit allowances, cross-checked against Artificial Analysis Intelligence Index v4.1.1 (re-verified Aug 17, 2026 via full-page curl ingestion of every AA model page and both pricing-plan pages — see footer for what's confirmed vs. inferred vs. unscored).
Brainstorm pick
—
—
Execution pick
—
—
Full task-cycles before budget runs out
—
—
01Weighting & tier thresholdⓘ
Value = intelligence ÷ blended $/M tok — higher is better. It measures the model itself, not a trade against the plan's top model, so a model's number never moves when other models are added or removed, and your value threshold means the same thing today as it will after the next catalog update. The old top-model "reference" concept is gone: the plan's smartest model competes like every other row and can win outright. A DOMINATED badge means some other model in the same table matches or beats it on both intelligence and price — it's mathematically never the better pick regardless of what its value reads, and it's excluded from winning; the number is shown for reference only. A FREE badge marks a $0/M tok model: its value is ∞ and it can win. Floor excludes models too far below the ceiling — brainstorm and execution each have their own floor slider, so tightening one has no effect on the other table. Execution tier threshold decides which credit allowance counts as "execution" for this plan — the brainstorm table below ranks every model regardless of tier, since a low-volume brainstorm pick doesn't need a small credit allowance, it just needs the best value. The value threshold below is separate: it is your own bar — the minimum intelligence per $/M tok a model must reach for Balanced mode to be satisfied. It only steers Balanced mode (Smartest/Cheapest ignore it by design), and it only prefers picks that clear the bar rather than hard-excluding anything (if nothing clears it, the best available value still wins rather than leaving no pick). The execution pick allowance below is separate: a cap on how far below the plan's top model the recommended execution model may sit. Since the brainstorm model usually bottlenecks the shared budget (section 06), a much-cheaper-but-much-dumber execution pick buys few extra cycles — e.g. on OpenCode Go with default floors, set the allowance to 10 and the algorithm is forced to pick within that band instead of the cheapest model with the best value. It only steers the algorithm pick in the execution table — a manual override in the ticker always wins — and it prefers within-allowance picks rather than hard-excluding anything (if nothing clears it, the best available pick still wins, flagged as outside your allowance).
15
15
∞
3.0×
5.0
02Brainstorm pick — all models, any tierⓘ
PICK current best pick in this table
ALGO PICK what the algorithm would pick, shown when you've overridden it below
FREE $0/M tok model — its value is ∞ and it can win
DOMINATED another model in this table matches or beats it on both intelligence and price — mathematically never the better pick, whatever its value reads
UNSCORED no published intelligence score yet
INFERRED score inherited from its base model, not independently benchmarked
CATALOG score comes from the plan's own catalog listing — no Artificial Analysis page exists
OVER YOUR THRESHOLD clears the floor, but sits below your value threshold — fewer intel points per $/M tok than you require (Balanced mode)
OUTSIDE ALLOWANCE the execution pick sits beyond your pick-allowance slider (best available, or your manual override) — execution table only
faded rows are excluded by this table's floor slider
Model
Intelligence
Δ Intel
Blended $/M tok
Δ Price
Intel / $/M tok
Credits/mo
Req/mo (published)
03Execution tierⓘ
Model
Intelligence
Δ Intel
Blended $/M tok
Δ Price
Intel / $/M tok
Credits/mo
Req/mo (published)
04Execution-tier runway for the job-hunting pipelineⓘ
Model
Intelligence
Credits/mo
$/run
Runs / 5hr · inferred
Runs / week · inferred
Runs / month
Total runs
Status at your input
Coverage
Model vs. model — every model in the plan, no tier restriction
05Session composition — brainstorm vs. execution shareⓘ
No public benchmark measures a "brainstorm phase vs execution phase" split directly — Artificial Analysis's Intelligence Index scores one-shot task completion, not a two-phase human-collaboration workflow. So instead of inventing a percentage, this derives the split from data that is published: exchanges × the picked brainstorm model's own request-token composition (input+cache+output, from the plan's published estimates) — or your manual override in the field below, against the pipeline-run token figure from section 04. That's the same per-task token-accounting approach AA uses on its model pages, applied to numbers these specific plans actually publish.
06Sequential capacity — tokens used when the budget runs outⓘ
This is a sequential capacity model: each cycle is one brainstorm phase (section 05 exchanges) followed by one execution run, back to back, so per-cycle spend is charged as a fraction of each model's own credit cap. The public docs don't specify how spend on one model affects what's left for another, so it's modeled as a proportional shared pool — each model's own credit cap represents 100% of the month, in its own terms — since that's how usage-based multi-model credit systems meter spend.
07Multi-run averaging — fixing the n=1 problemⓘ
Add token counts from your actual past pipeline runs below. Once you've got more than one, "Use as pipeline-run tokens" pushes the average into section 04 (and section 05) in place of the single 207k estimate.