Choose a model for the work.

Choose an AI model for marketing or sales. Compare a sourced shortlist, estimate API token costs and export a practical evaluation brief. Free, no sign-in.

20 models across OpenAI, Anthropic, Google, Moonshot, xAI, DeepSeek, Mistral and Z.ai. Official sources checked 2026-11-13; review due 2026-11-20. This is an editorial shortlist, not a head-to-head benchmark or a live provider feed.

Start with the marketing job

  • Campaign copy & content briefs: A general writing model is a sensible starting point when it must preserve product facts and follow a brand brief. Reject invented claims or a mismatched offer. Have a reviewer choose the clearer, more specific draft without seeing the model name.
  • Sales follow-ups & account briefs: The work needs accurate use of supplied account facts and a clear next action. Fluent personalisation alone is insufficient. No invented relationship, promise or commitment. The unresolved point stays unresolved; a named human approves sending.
  • Lead tagging & CRM extraction: A bounded taxonomy with checkable fields is a good place to test economical models. Measure per-label errors and abstentions. A wrong high-impact route is a failure even when aggregate accuracy looks good.
  • Competitor & market research: Useful research depends on a working evidence-collection process as well as synthesis. Every material claim traces to an inspected source. Missing data stays unknown. Implications are clearly distinguished from observations.
  • Marketing analysis & reporting: General models can explain an analysis; executed calculations must supply the numbers. Visits −20%; enquiries −10%; rate 4% to 4.5%, +0.5 percentage points. No causal claim follows from those counts alone.
  • Positioning & difficult trade-offs: Conflicting evidence and costly decisions justify a separate reasoning shortlist and closer human review. The memo exposes the decisive uncertainty, an affordable next test and a stop condition. Reject invented market sizes and certainty scores.
  • Landing-page & screenshot review: This shortlist has documented image input. The review still needs actual screenshots and the intended user action. Reject invented controls and generic advice. Verify the proposed changes on the page with keyboard and mobile interaction checks.
  • Interview audio & video analysis: Gemini 3.8 Flash is the directly verified audio/video-input candidate in this curated catalogue. One result is not a competitive winner. Check timestamps and speaker attribution. The later correction must replace the earlier claim without mixing products.
  • Marketing tools, SQL & integrations: Use a coding-capable candidate, with completion judged by the delivered behaviour and tests. Run tests for duplicates, invalid inputs and timeouts after a possible write. Check mobile behaviour and permissions before deployment.
  • Research public conversations on X: Grok with X Search is a specific research route. The base model alone does not provide live social evidence. Verify source links and avoid treating a convenience sample as market prevalence. No automated engagement is implied.

API cost method

Cost = requests × (billable input tokens × input rate + billable output tokens × output rate) / 1,000,000. Include billed reasoning in output. Compare short uncached standard text rates, with conservative peak prices for DeepSeek. Media, tools, retries, taxes, subscriptions and human review are excluded. The interactive calculator accepts at most 100,000 input and 8,000 output tokens per request. Chat mode uses editorial comparison order and does not rank subscription value.

Worked example

At 2,000 input and 500 output tokens, GPT-6.1 Sol costs $0.009 per request, or $9 for 1,000 requests, before extras. Equal assumed tokens are a planning convention, not measured model usage. A lower token bill does not prove better accepted-output economics.

Source catalogue

USD per million text tokens, snapshot of 2026-11-13. Dated changes are shown below. Enable JavaScript to filter, calculate and export your evaluation brief. No model calls, sign-in or API credits are required.

How to decide

Keep the same approved brief and acceptance criteria for two candidates. Track failed attempts and human review. Use the least expensive workflow that meets your standard. Availability, data terms and exact model access require checking with the provider. Open-weight licences do not make hosting free.

Read the full model-selection guide