CalcRecipe

AI Decision Platform

KI-Stack-Planner

Turn product type and constraints into a scored shortlist from the shared model registry.

Product 2026.07.5 · Catalog 2026.07.2 · Pricing pricing-2026.07.2 · Trust 2026.07.3 · Decision 2026.07.4 · Core 1.0.0

Product constraints

Custom pricing applies to matching models in the comparison. Every row uses the shared pricing resolver.

Preisbasis

Preisquelle: Nicht verfügbar

Nicht verfügbar

Token pricing is not verified for this model. API availability may be confirmed, but cost calculations require custom pricing. Missing prices are never treated as $0.

API evidence: owner_api_account · 2026-07-20 · account-scoped

Limits (not pricing): TPM 500000 · RPM 500 · TPD 900000

Last verified: —

Dataset version: pricing-unavailable

Verification date unknown — treat this pricing as provisional.

Einige Preisdaten können unvollständig oder veraltet sein. Dieses Ergebnis ist eine Planungsschätzung auf Basis der von CalcRecipe verifizierten Preise. Die tatsächliche Abrechnung kann abweichen. Bestätigen Sie finale Preise beim Anbieter.

Recommended stack

Recommended model

Gemini 2.5 Flash

Est. monthly AI cost

361,20 $

Fit score

84

Entscheidungskonfidenz

53

Budget fit

Within budget

Why this stack

  • • Strong on Cost efficiency (weighted 0.179 of total 0.839).
  • • Strong on Budget fit (weighted 0.150 of total 0.839).
  • • Strong on Context window (weighted 0.120 of total 0.839).
  • • Tier: budget
  • • Pricing source: verified

Warum diese Empfehlung

Entscheidungskonfidenz: 53/100 (low)

Hauptgründe

  • • Strong on Cost efficiency (weighted 0.179 of total 0.839).
  • • Strong on Budget fit (weighted 0.150 of total 0.839).
  • • Strong on Context window (weighted 0.120 of total 0.839).
  • • High latency priority favors budget-tier models.

Why it won

  • • Strong on Cost efficiency (weighted 0.179 of total 0.839).
  • • Strong on Budget fit (weighted 0.150 of total 0.839).
  • • Strong on Context window (weighted 0.120 of total 0.839).
  • • Policy emphasized: Cost efficiency 20%, Quality / capability tier 18%, Budget fit 15%.
  • • High latency priority favors budget-tier models.

Alternativen

  • GPT-4o mini

    • • Total score 0.748 vs winner 0.839 (gap 0.091).
    • • Trails on Context window (weighted 0.008 vs 0.120).
    • • High latency priority favors budget-tier models.

  • Gemini 2.5 Pro

    • • Total score 0.678 vs winner 0.839 (gap 0.161).
    • • Trails on Cost efficiency (weighted 0.088 vs 0.179).
    • • Trails on Budget fit (weighted 0.000 vs 0.150).

  • DeepSeek Chat

    • • Total score 0.653 vs winner 0.839 (gap 0.186).
    • • Trails on Context window (weighted 0.000 vs 0.120).
    • • Trails on Vision (weighted 0.000 vs 0.080).

Removed by constraints

  • • gpt-5: Pricing unavailable for this model
  • • gpt-5-codex: Pricing unavailable for this model
  • • gpt-5-mini: Pricing unavailable for this model
  • • gpt-5-nano: Pricing unavailable for this model
  • • gpt-5-pro: Pricing unavailable for this model
  • • gpt-5.1: Pricing unavailable for this model
  • • gpt-5.1-chat-latest: Pricing unavailable for this model
  • • gpt-5.1-codex: Pricing unavailable for this model

Warnungen

  • • Some candidates use estimated pricing.
  • • Some pricing data may be stale.

How confidence was determined

  • • Pricing confidence for the winner is solid.
  • • Estimated catalog pricing is in play.
  • • Stale pricing reduces decision confidence.
  • • Clear score gap between winner and runner-up.
  • • Many candidates were removed by hard constraints.
  • • Sensitivity analysis shows the recommendation flips under multiple scenarios.

Sensitivität

  • Higher cost priority: Recommendation stays Gemini 2.5 Flash under “Higher cost priority”.
  • Higher quality priority: Recommendation changes to Gemini 2.5 Pro under “Higher quality priority”.
  • Lower budget (−30%): Recommendation changes to GPT-4o mini under “Lower budget (−30%)”.
  • More requests (+50% users): Recommendation changes to GPT-4o mini under “More requests (+50% users)”.
  • Higher latency tolerance: Recommendation stays Gemini 2.5 Flash under “Higher latency tolerance”.
Policy:
ai-stack-policy-2026.07.1
Catalog:
2026.07.2
Pricing dataset:
pricing-unavailable
Calculator:
1.1.0
Calculated at:
2026-07-21T03:54:03.410Z

Alternatives & cost comparison

  • Gemini 2.5 Flash

    score 84

    Est. monthly AI cost: 361,20 $

  • GPT-4o mini

    score 75

    Est. monthly AI cost: 104,40 $

  • Gemini 2.5 Pro

    score 68 · over budget

    Est. monthly AI cost: 1.455,12 $

  • DeepSeek Chat

    score 65

    Est. monthly AI cost: 187,20 $

  • GPT-4o

    score 54 · over budget

    Est. monthly AI cost: 1.740,00 $

Assumptions

  • • Baseline workload: 20 requests/user/day, 800 input / 400 output tokens, 20% cache.
  • • Ranking uses the Decision Engine Core with a versioned AI Stack policy — not hidden heuristics.
  • • Hard constraints remove invalid models; soft preferences adjust rank with explicit reasons.
  • • 20 requests/user/day
  • • 800 input / 400 output tokens
  • • 20% cached requests
  • • Product type: chat
  • • Monthly budget: $400
  • • Priority preset: balanced

Planning estimate

Dieses Ergebnis ist eine Planungsschätzung auf Basis der von CalcRecipe verifizierten Preise. Die tatsächliche Abrechnung kann abweichen. Bestätigen Sie finale Preise beim Anbieter.

Examples

  • RAG docs assistant: Prioritize quality + large context → Gemini Flash/Pro or Claude long-context tiers.
  • High-volume chat: Raise latency priority and tighten budget → mini/Haiku/DeepSeek shortlist.

Frequently asked questions

Is this a provider affiliation?
No. Recommendations are heuristic scores over an open configuration catalog for planning.
Why can provider pricing change?
Providers update list prices, promotions, batch rates, cache rates, long-context tiers, and enterprise contracts without notice. CalcRecipe labels freshness so you can decide when to re-check.
When should I use custom pricing?
Use custom pricing when your invoice, aggregator markup, or negotiated contract differs from the public catalog. Custom inputs are labeled User Override and never shown as Verified.
How does CalcRecipe label pricing sources?
Verified, Estimated, Custom, Stale, or Unavailable. The source of every calculation stays visible in results and share snapshots.
Should I trust these numbers for contracts?
No. Always confirm official billing terms with the provider before financial commitments.

Choosing an AI stack with constraints

Product type filters capability tags; budget and priority sliders reshape scores.

Always pair stack choice with the Product Cost Planner using your real traffic shape.

Provider list prices change often and may differ by region, channel, promotions, batch/cache rates, or enterprise contracts. CalcRecipe separates verified catalog rates from estimates and user overrides.

Custom pricing is browser or session state only—it is not public content and never silently replaces CalcRecipe Verified data.

Version history

  • 2026.07.4

    2026-07-20

    Introduced the domain-agnostic Decision Engine Core; migrated AI Stack Planner to explainable weighted scoring.

    Builder recommendations must be reproducible, constraint-aware, and explainable across domains.

  • 2026.07.3

    2026-07-20

    Added the AI Pricing Trust Layer with user overrides, freshness states, and transparent source labels.

    Catalog pricing can be incomplete or stale; builders must inject real rates without mislabeling them as verified.

  • 1.1.0

    2026-07-20

    AI Stack Planner v1.1 — Decision Engine adapter with priority presets, confidence, sensitivity, and policy snapshots.

    Replace opaque heuristic scores with a versioned decision policy.

  • 1.0.0

    2026-07-20

    AI Stack Planner v1 — product-type aware model recommendations.

    Reduce stack selection to budget, latency, and quality priorities.

Related AI planners

CalcRecipe Werkstatt-Leitfäden