The models we use

The right model for the job.

New AI models ship every month. We track them, test them, and put the right one behind each part of your business, so you get the outcome without keeping up with it yourself.

Updated 21 November 2026

By job

Which model does which job.

Business taskModelWhy
Answering and qualifying enquiries, 24/7Fast, low-cost modelse.g. Claude Haiku 5.5, GPT-6 LunaInstant replies at a low running cost, day and night.
Proposals, analysis and complex decisionsFrontier reasoning modelse.g. Claude Fable 5.2, GPT-6 Astra, Gemini 4 CarbonThey can follow multi-step business logic and weigh trade-offs.
Long documents, reports and contractsLong-context modelse.g. Claude Opus 5.5, Claude Sonnet 5.5They read and draft from the full document, such as a property valuation report, not a summary of it.
Building and maintaining your softwareCoding and agentic modelse.g. Claude Opus 5.5, GPT-6.1 Sol, Gemini 4 HydrogenFaster, cheaper builds and changes, checked by a person before they ship.
Voice and phoneSpeech modelse.g. GPT-Live 1, Gemini 3.8 LiveCalls answered in a natural voice, then summarised for your team.
Reading forms, photos and invoicesVision modelse.g. GPT-6.1 Terra, Gemini 3.8 Flash, Claude Sonnet 5.5The details come out of the image, so nobody retypes them.

Running cost

Cost to answer 1,000 customer enquiries.

Estimate In US dollars, at each company’s published standard API price.

Estimated cost in US dollars to answer 1,000 customer enquiries, by model
ModelCompanyEstimated cost (USD)
GPT-6 LunaOpenAI$0.55
Gemini 3.8 FlashGoogle$4.13
GPT-6.1 TerraOpenAI$5.50
Gemini 4 HydrogenGoogle$8.25
Claude Sonnet 5.5Anthropic$11
GPT-6.1 SolOpenAI$11
Gemini 4 CarbonGoogle$11
Gemini 4 ArgonGoogle$11
Gemini 4 NeonGoogle$17
Claude Opus 5.5Anthropic$22
Claude Fable 5.2Anthropic$44
GPT-6 AstraOpenAI$55
Method: each enquiry is counted as about 3,000 tokens in (your instructions, business details and the conversation) and 500 tokens out, multiplied by each company’s official per-token price, checked 21 November 2026. Excludes caching discounts, tax and hosting. Voice, transcription and open-weight models are priced differently and aren’t shown.

What’s one extra job worth?

Type your numbers to see the comparison.

Running cost uses GPT-6.1 Sol, the middle of the chart above, at its official price checked 21 November 2026, and the same enquiry size as the chart. It’s an estimate of the AI’s cost only.

The AI itself costs little to run. Most of the cost is building and running the system around it. We quote that after a 60-minute consultation.

Talk to AETHER

Current models

The models we work with now.

Checked every day against each company’s own announcements, documentation and changelogs. New generally available models are added automatically from those sources only.

OpenAI

Anthropic (Claude)

Google (Gemini)

Beyond the model

What we connect the model to.

We connect it to your CRM, email, calendar, database and tools. We give it skills, which are repeatable procedures, and your business rules. And we keep a person in the loop for anything that matters.

When a better model ships, we test it on your real work and switch if it’s better, so you’re never locked into one company’s model.

Start here

What should be working better?