Skip to content

AI models: choose a tool for the task

A 5 September 2026 reference edition for comparing text, image and video models, with primary sources and a practical trial.

By The Leverage Club. Sources checked .

Free to read, no account needed. Tool access, prices and terms can change; check the linked sources before a run.

Start with the task, then compare

This is a historical snapshot checked on 5 September 2026, not a live model ranking. Later releases may replace or change these entries. Check the primary source and your account before relying on a model name, status or feature.

Choose three representative tasks, use equivalent inputs and count corrections as part of the cost. Keep the workflow you can verify. The club has not run a comparative benchmark of these models.

Run the model comparison exercise to turn these notes into a decision you can explain.

For the later image exercises, see the 8 September product-editing lab. The Suno v6 guide has its own 13 September source check.

GPT-6 Astra

Reasoning, coding and research. Status in this edition: Check account access.

OpenAI positions Astra for difficult work spanning tools and documents. It accepts text and images and produces text; image generation is a separate tool. A sensible candidate for a demanding brief or a code review.

What to check

Compare it on the same task and inputs as your current model. Record corrections and total cost, including tools. Do not assume it is available in every plan or in Cirka.

Read the official GPT-6 Astra reference

Claude Fable 5.1

Long projects and knowledge work. Status in this edition: Generally available, per Anthropic.

Anthropic's September 1 release targets coding and extended knowledge work. Its product page lists paid Claude plans and API access. Consider it for a second assessment of a complex plan, not an automatic replacement for a cheaper model.

What to check

Check permissions, retention and the current price before uploading client material. Our course has not run a comparative benchmark.

Read the official Claude Fable 5.1 reference

Gemini 3.8 Flash

Agent workflows and coding. Status in this edition: Listed as stable.

Google's model catalogue identifies 3.8 Flash as its current Flash model for complex workflows. Use the exact model identifier in an experiment log so a future alias change does not erase what you tested.

What to check

Start with a small representative workload. Measure task success and latency instead of inferring either from the word Flash.

Read the official Gemini 3.8 Flash reference

GPT Image 2

Image generation and editing. Status in this edition: Earlier generation · 5 Sep notes.

OpenAI documents text-and-image inputs and generated images. Use it as one candidate for product-preservation and layout tests, with the same reference photo across candidates.

What to check

Inspect small labels, geometry and reflections. A visually attractive result is not proof that the product remained accurate.

Read the official GPT Image 2 reference

Nano Banana 2.1 / Nano Banana Pro

Image generation and editing. Status in this edition: 2.1 generally available since 6 Oct 2026; Nano Banana 2 deprecated.

Google lists Nano Banana 2.1 as gemini-nano-banana-2.1 and Pro as gemini-3-pro-image. They are distinct choices; write down which one you used instead of logging only Nano Banana.

What to check

Compare cost per accepted image. Check the engine actually offered by your chosen app before following a model-specific instruction.

Read the official Nano Banana 2.1 / Nano Banana Pro reference

Runway Gen-4.5 / Aleph 2.0

Video generation / video editing. Status in this edition: Documented video models.

Runway documents Gen-4.5 and Aleph 2.0 as separate video tools. Choose a generation workflow for a new shot and an editing workflow when you need to modify existing footage.

What to check

Check the endpoint, duration and supported input format. Score usable seconds, continuity and editing effort; do not mix text-to-video and image-to-video results in one ranking.

Read the official Runway Gen-4.5 / Aleph 2.0 reference

Gemini Omni Flash

Video and native audio. Status in this edition: Listed as preview.

Google lists gemini-omni-1.1-flash for video generation and editing with audio. Preview status makes this an experiment candidate, with a fallback for client deadlines.

What to check

Test the exact mode you need. Keep the original stills, script and edit timeline so you can change tools without rebuilding the entire project.

Read the official Gemini Omni Flash reference

Claude Mythos 5.1

Research watchlist. Status in this edition: Restricted access.

Anthropic limits Mythos 5.1 to vetted organizations. An announcement does not make it a practical course prerequisite.

What to check

Follow the release for context; use an available model for the exercises. Do not buy access from unofficial resellers.

Read the official Claude Mythos 5.1 reference

Check the workspace as well as the model

Cirka is our founder's app. We use it as an optional creative workspace in these examples. A separate subscription or credits may be required. Check the tools and models actually available in your account; a model mentioned in a guide is not a claim of Cirka integration.

Explore Cirka Creative Studio when a reviewed brief is ready for image production. Preparing and comparing the brief does not require a Cirka subscription.

Continue from here

See how this exercise fits into the Vibe coding course, including its full public syllabus.

Browse all free guides for another task, or save your work and return to it later.

Keep learning in the club

Course outlines and these guides are free to read without an account. Full lesson bodies and member features require a paid membership. The preview uses sample data; external tools may charge separately.