Who is behind the guide
Marc Mojica publishes this guide. AI assists research, drafting, and site development. Recommendations are editorial starting points; they are not a universal leaderboard or a claim that every listed model has been benchmarked by us.
How we make recommendations
- Start with the task. Match documented capabilities, available access, cost, and operational constraints to a concrete job.
- Use original sources. Model makers establish specifications and lifecycle dates. Platform documentation establishes availability on that platform. A maker’s model card alone does not prove every cloud hosts it.
- Keep facts separate from judgment. File sizes and published prices are sourced facts. Suggested uses, memory allowances, and tutorial timings are labeled estimates or editorial guidance.
- Make the final choice with examples. Test representative inputs, record quality and corrections, and measure total cost and latency on your own workload.
Review dates and changing prices
The pricing shortlist and detailed availability records were reviewed September 7, 2026. Older family and platform summaries retain their August 9 review date. A new site release does not make every old record newly verified.
Catalog family and variant records older than 30 days are marked for review. Known promotion transitions update on their effective date. If a promotion ends without an established replacement rate, it is excluded from current cost calculations until rechecked. Model retirement notices are shown beside affected variants.
Prices are USD standard paid API rates per million text tokens unless stated otherwise. Calculations account for documented long-context tiers, but exclude caching, tools, media, retries, taxes, and regional differences. Consumer subscriptions are a separate purchase.
What the assistant can establish
Ask LLM.how uses selected catalog records and their approved source links. Cost comparisons calculate from our dated price shortlist and identify their scope. Selected starter questions use clearly labeled, reviewed reference comparisons. Generated guidance may still be wrong: inspect the linked sources and test the recommendation. When an answer cannot be generated or verified, the assistant offers a clearly labeled catalog fallback.
Tutorials and local memory estimates
Tutorial sample organizations, source packets, meetings, and answer keys are fictional. They let you check a workflow without uploading work documents. Memory tiers assume one short local chat and are planning estimates, not measured minimums or a guarantee of speed.
Suggest a correction
Include the page URL, the statement or step to change, an official source or reproducible example, and the date you checked it. Please omit credentials and private work material.
Contact Marc ↗September 8 creative update
Added dedicated video and voice and audio guides, with MiniMax H3, Seedance 2.5, MiniMax Speech 2.8, Sonic 3.6, current ElevenLabs speech and transcription, and MiniMax Music 3. These source-linked records were reviewed September 8 and also power catalog search and assistant context. Other records retain their own review dates.
September 7 changes
- Updated the pricing shortlist and removed the cancelled Sonnet price increase.
- Added exact variant availability, lifecycle notices, and local memory estimates.
- Improved catalog search, comparisons, and assistant recovery.
- Published three complete tutorials with reusable sample workbooks.