Skip to content
How fit is your AI?Take the quiz
Sign inOptimize your work

Orchestra vs Not Diamond

Not Diamond
A routing engine that recommends which model to call. You run the call through your own gateway and provider keys.6,7
Orchestra
Measures every workflow against your definition of success, improves each one, and trains models your team owns.9

Side by side

Orchestra and Not Diamond compared on 4 dimensions
Primary jobNot DiamondPredicts which model to use for each prompt, aiming for frontier quality at a lower price. The homepage and pricing now center on coding agents.1,2OrchestraEvaluates production work, improves the prompts and routes behind it, and trains specialist models your team owns.9
Gateway and model accessNot DiamondReturns the recommended model, then you call it with your own SDK and an API key for each provider. Not Diamond does not supply the models.6,7OrchestraOrchestra is the gateway. Your existing OpenAI or Anthropic SDK calls it with an Orchestra API key, and Orchestra handles model access, routing, and measurement.9
Evaluation approachNot DiamondCustom routers train on your evaluation data: prompts, each candidate model's responses, and scores from any metric you choose. Prompt optimization scores prompts on your examples and measures them on a held-out test set.3,4OrchestraBuilds task-specific benchmarks, such as whether the model updated the right record or followed your policy, and compares models on the same held-out work.9
Model improvementNot DiamondTrains routers on your data and rewrites prompts for each target model.3,5OrchestraRepairs prompts, schemas, parsers, and tool contracts first. For a recurring task, trains a specialist from accepted traces and expert corrections.9

Not Diamond is one piece.Orchestra is the whole stack.

Not Diamond decides which model should answer. Everything around that decision is still yours to build and run.

What you assemble around Not Diamond

  • A gateway or SDK that makes the call to the recommended model.6,7
  • Accounts and API keys with each model provider you route between.6
  • Scored responses from each candidate model, if you want a custom router.3
  • Checks that tell you whether the answer was right for the task.

What Orchestra includes

  • The gateway your existing OpenAI or Anthropic SDK already calls.9
  • Model access, routing, and security in the same place.
  • Did the agent update the right record?9
  • Specialist models trained from accepted traces and expert corrections.9

Where Orchestra is different

  1. One system instead of parts

    With Not Diamond, your team still runs the gateway, holds the provider keys, and builds the checks. Orchestra runs the gateway, the routing, and the measurement together.6,9

  2. Quality defined by your work

    Generic benchmarks don't show how a model performs on your task. Orchestra writes the examples, rubrics, and checks that define success for each workflow.9

  3. A specialist model your team owns

    For a recurring task, Orchestra trains a specialist from accepted traces and expert corrections.9

When each approach makes sense

  • Not Diamond makes sense when

    • You already run your own gateway and provider accounts and only need a routing decision.6,7
    • You have scored responses from several models and want a router trained on them.3

  • Orchestra makes sense when

    • You want the gateway, model access, routing, and measurement from one place.
    • The workflow runs in production and your business depends on it.
    • You know what you want from the work, and every change is checked against that goal on your own examples.9
    • You want to improve prompts and routes, then train a specialist you own.9

  • Both can be part of the stack when

    • You already use Not Diamond for routing. Orchestra works with it and adds task-level measurement.

Orchestra case studies

  • Orchestra case study

    Teaching a CRM agent to finish the job

    75.5%

    lower model cost than Sonnet 4.6

    Model spend per seven-task AutomationBench CRM run: $0.27 for Qwen 3.6 Plus with a revised prompt, $1.12 for Sonnet 4.6. The revised prompt also scored 13.1% higher on mean task score.

    Read the story
  • Orchestra case study

    Cutting response times for everyday operations

    5.2×

    faster median model response than Sonnet 4.6

    The small model passed every task check on 92.2% of attempts; Sonnet passed 100%. Both ran 30 AutomationBench workflows, with 90 attempts per model.

    Read the story
  • Orchestra case study

    Finding purchase intent in 40,000 comments

    $2.82

    model cost for 39,962 comments

    An Orchestra post-trained 30B open model labeled 39,962 YouTube comments from a Snowflake table and returned 39,949 valid outputs. Sonnet cost $12.48 on the same table.

    Read the story

These are Orchestra experiments that compare models on specific workloads. None of them compare Orchestra with Not Diamond.

Sources (12)Show

Sources reviewed October 2026. Not Diamond statements come from its own website and documentation. If something is not described there, we say so instead of assuming it is missing.

  1. 1.Not Diamond homepage notdiamond.ai (opens in a new tab)
  2. 2.Not Diamond: Pricing and FAQ www.notdiamond.ai/pricing (opens in a new tab)
  3. 3.Not Diamond docs: Training a custom router docs.notdiamond.ai/docs/router-training-quickstart (opens in a new tab)
  4. 4.Not Diamond API reference: Prompt optimization docs.notdiamond.ai/reference/optimize_prompt_v2_prompt_optimize_post (opens in a new tab)
  5. 5.Not Diamond docs: What is prompt optimization? docs.notdiamond.ai/docs/what-is-prompt-optimization (opens in a new tab)
  6. 6.Not Diamond docs: Pre-trained router quickstart docs.notdiamond.ai/docs/quickstart-routing (opens in a new tab)
  7. 7.Not Diamond API reference: /modelSelect docs.notdiamond.ai/reference/token_model_select_v2_modelrouter_modelselect_post (opens in a new tab)
  8. 8.Orchestra homepage
  9. 9.Orchestra platform
  10. 10.Case study: Teaching a CRM agent to finish the job
  11. 11.Case study: Cutting response times for everyday operations
  12. 12.Case study: Finding purchase intent in 40,000 comments

Bring one workflow. See how it scores.

Pick a recurring task. We'll define what good looks like and compare models on your own examples.

Evaluate your workflowsSee other comparisons