Upfyn Rishi: Hard multi-step reasoning, research, media

Rishi, the Upfyn deep · agentic model
Deep · Agentic

Meet Rishi

Best for hard multi-step reasoning, research, media.

Rishi plans, executes, and then verifies its own work across long multi-step runs, which is why it is slower and why it costs the most per token. It is the only model that accepts video as input and generates video. Reach for it when being wrong is expensive — research you intend to act on, a migration you are about to run, an analysis someone will make a decision from — and let the other two carry everything else.

Deep · AgenticModels

Listen to article

In short

  • Plans, executes, and then verifies its own work
  • Holds long multi-step runs together, end to end
  • The only model that accepts video as input and generates video
  • A 1M-token context
  • Slower and the most thorough — for when being wrong is expensive

Plans, executes, then checks

Rishi works across long multi-step runs: it plans the work, carries it out, and verifies the result before you see it. That is why it is slower, and why it costs the most per token.

1 / 10

Research you intend to act on

Reach for Rishi when a plausible wrong answer is the failure: research you mean to act on, an analysis someone will decide from, a change you are about to run. Let Bala and Yuva carry everything else.

Video in, video out

Rishi is the only one of the three that understands video and generates it. Video generation is available through /v1/videos.

Finish your research with something you can use

Compare options, collect evidence, and turn research into a useful brief with Upfyn. Keep the findings connected to the sources behind them.

1 / 10

Give your weekly report a repeatable starting point

Build a weekly report from your working files, refine it in chat, and schedule the preparation so you can spend more time reviewing the result.

Where it runs

Rishi is available wherever you use Upfyn: the desktop app, the web app and your phone, and through the UpfynAI gateway under the model id upfyn-rishi. It is one of Upfyn’s own frontier models, hosted on AWS in the India region. Your projects, tools and permissions stay with you when you switch models.

  • Chat/v1/chat/completions
  • Images/v1/images/generations
  • Speech/v1/audio/speech
  • Transcription/v1/audio/transcriptions
  • Video generation/v1/videos

How the three compare

BalaYuvaRishi
Best forEveryday chat, drafting, high-volume textProduction chat, coding, vision + audioHard multi-step reasoning, research, media
SpeedFastestFastDeliberate
DepthDirect — single passConsidered — up to 2 stagesAgentic — most thorough
ModalityTextText + image + audioText + image + audio + video generation
Contextup to 1M1M1M
PlansEvery plan, including FreeEvery plan, including FreeEvery plan · usage charged in credits

Where it stops

Longer reasoning and media generation use more of your usage limit. Usage depends on the request, output length and tools used; there is no fixed number of runs per plan.

Availability

Every plan · usage charged in credits. Usage counts toward your plan’s limit; the plans and what each includes are on the pricing page.

Specifications

Model id
upfyn-rishi
Speed
Deliberate
Depth
Agentic — most thorough
Modality
Text + image + audio + video generation
Context
1M (1,000,000 tokens)
Plans
Every plan · usage charged in credits
Hosting
AWS, India region

The other two

Keep reading