Supported models

Every frontier lab. One employee.

Jyper is not built on one AI lab. It runs the current frontier models from all of them and sends each task to the one that does that job best. When a better model ships, Jyper switches. You don't.

AnthropicOpenAIGoogleDeepSeekMoonshot AIAlibabaZ.ai
The roster

What's under the hood today

You never pick a model. This is simply what Jyper has to choose from, and what each one is currently best at.

Anthropic

Deepest reasoning, where judgment carries the file.

The benchmark deep dive →

Claude Opus 5

Reserved for the hardest costings: multi-leg, multi-currency, margin rules full of exceptions.

Deep reasoning1M contextVision

Claude Opus 4.8

Long autonomous runs. The overnight batch that still has to hold together at 6am.

Reasoning1M contextVision

Claude Sonnet 5

Near-Opus judgment at speed. The everyday workhorse behind most drafting.

Reasoning1M contextVision
OpenAI

Broad frontier strength across the whole file.

The benchmark deep dive →

GPT-5.6 Sol

OpenAI’s flagship. A second strong opinion on the requests worth winning.

Deep reasoningVision

GPT-5.6 Terra

The balanced tier: revisions, structured extraction, first-pass review.

ReasoningVision

GPT-5.6 Luna

Cheap and quick, for the thousands of small reads behind one quotation.

Speed
Google

The long window. Documents, images, memory.

The benchmark deep dive →

Gemini 3.7 Flash

Frontier speed on a million-token window. Rate sheets and long email threads, read at pace.

Speed1M contextVision

Gemini 3.1 Pro

The long-document specialist. A full supplier contract read end to end: seasons, supplements, blackout dates.

Reasoning1M contextVision

Nano Banana 2

Generates and adapts the visuals that make a proposal read like a brochure, not a spreadsheet.

Image generation

Gemini Embedding

The memory. Finds the right supplier, rate or past trip across everything you have loaded in.

Search & memory
DeepSeek

Open weights, built for tool use.

The benchmark deep dive →

DeepSeek V4-Pro

Strong on long, multi-step work where the model has to keep using tools to get to an answer.

Deep reasoningAgentic

DeepSeek V4-Flash

The cheap lane. Routine transformations where turnaround matters more than depth.

Speed
Moonshot AI

Stamina across long agentic sessions.

The benchmark deep dive →

Kimi K3

Holds its thread through dozens of small tasks back to back, which is what supplier chasing actually is.

AgenticReasoningVision
Alibaba

Vision at scale.

The benchmark deep dive →

Qwen3.8-Max

For the visual side of travel ops: scanned contracts, photographed rate cards, screenshots of availability.

Vision1M context
Z.ai

Fast and cheap where depth is not the point.

The benchmark deep dive →

GLM-5.2

Quick drafts and quick transformations, at a fraction of frontier cost.

Speed
Why it is built this way

Model lock is the
new vendor lock.

If your tool only runs one lab, you gain nothing the week another lab jumps ahead. And nobody knows who will be ahead next year. Because Jyper runs them all, every upgrade reaches you without a migration, a retraining or a new contract.

The routing

A quotation is many jobs. Each gets its own specialist.

Reading a brief, chasing suppliers, holding a six-day itinerary together and writing the proposal reward completely different strengths. No single model wins all of them, so Jyper assigns each job to the model that does, based on benchmarking across thousands of real RFPs.

Reading the brief

Careful extraction from a long, messy email chain.

Gemini 3.7 Flash · Claude Sonnet 5

Supplier discovery

Wide, fast reading across your network and the open web.

Gemini 3.7 Flash · GPT-5.6 Luna

Contracts & rate sheets

A very long context window and real vision.

Gemini 3.1 Pro · Qwen3.8-Max

Outreach & negotiation drafts

Judgment and tone, written under your rules.

Claude Opus 5 · GPT-5.6 Sol

Itinerary & logistics

Long-horizon reasoning that keeps day six consistent with day two.

Claude Opus 5 · GPT-5.6 Sol

Overnight chasing

Stamina across dozens of small tool-using steps.

Kimi K3 · Claude Opus 4.8

Costing sheets

Precision. Models read and structure; every number is computed by Jyper’s own pricing tools.

Claude Opus 5 · DeepSeek V4-Pro

Proposal writing

Whoever writes best that week. Proposals are won on the page.

Rotates with the roster

Search across your history

Recall over everything your organisation has ever loaded in.

Gemini Embedding
Upgrades, not migrations

A model joins the roster when it beats the current lineup on real DMC work, and leaves when something better ships. The swap happens under the hood. Nothing changes on your side, and there is no version of Jyper you have to move to.

Where models are not trusted

Models read, structure and write. They do not invent numbers. Every rate, margin and total in a proposal is computed by Jyper's own pricing tools and traced back to the contract it came from, whichever model did the writing.

Roster last reviewed August 2026. Frontier models ship faster than websites get updated, so treat this as the current state rather than a fixed list.

The interview

Meet your
next employee.

Thirty minutes, on a call. Bring a real request from your inbox and watch Jyper hand back a costed proposal — live. If it isn't the strongest operator you've ever interviewed, don't hire it.

One DMC per destination · the best time to hire is before your season.

or write to us — hello@jyper.io