GLM-4.6 Exacto: What It Is, API Routing, and Workload Fit

If you mean GLM-4.6 with OpenRouter’s :exacto suffix, Exacto is a routing variant—not a separately trained GLM model. OpenRouter describes it as quality-first provider routing that favors providers with stronger tool-calling reliability signals. That may suit tool-using agents, but it does not guarantee successful tool calls or a particular provider for every request. (Exacto documentation, model-variant documentation)

What Exacto changes

OpenRouter documents :exacto as a routing suffix. Its model-variant documentation says routing variants can be added to model IDs at request time. Applied to the GLM-4.6 model listed in OpenRouter’s catalog, the selector is z-ai/glm-4.6:exacto. Exacto changes provider routing; it does not change GLM-4.6’s weights or capabilities. (Exacto documentation, model-variant documentation, GLM-4.6 catalog)

Consider Exacto when your workflow depends on tool calls and you’re willing to prioritize tool-use quality signals over lowest-cost routing. For non-agentic work where price is the main concern, compare default routing too. Exacto is a routing preference—not a guarantee of tool-call accuracy, a fixed provider, or lower cost. (Exacto documentation)

Is it a fit for your workload?

Exacto is worth evaluating for agentic tasks where tool selection and valid arguments matter—for example, an assistant that calls search or database tools, or a coding agent that calls tools during a task. That recommendation reflects OpenRouter’s stated purpose for Exacto, not demonstrated improvement on every GLM-4.6 task. (Exacto documentation)

For general GLM-4.6 limits, Z.ai documents a 200K-token context and 128K maximum output. These are base-model specifications, not evidence of Exacto-specific performance. Confirm the limits and parameters applicable to your chosen route and request. (Z.ai GLM-4.6 migration guide)

A Cline article reports a 94.9% GLM-4.6 success rate on its diff-edit telemetry, compared with 96.2% for Claude Sonnet 4.5. Those are Cline-reported results; the article does not give enough evaluation detail to predict performance on your codebase. The figures do not measure Exacto routing. Treat them as context, not a production threshold. (Cline report)

Keep another option or a fallback available if your workload is mostly non-agentic, cost is the overriding concern, or tool failures would be costly. Validate tool outputs and handle errors in your application; routing preferences do not replace those controls.

Production API: selector, endpoint, and cost

OpenRouter’s Exacto documentation shows a Responses API request with model and string-valued input, sent using POST to /api/v1/responses with a bearer API key. Its example uses a different model. The following is a minimal adaptation: the endpoint, method, headers, and body shape follow that documented example; the GLM-4.6 selector applies OpenRouter’s documented routing-variant rule to its listed GLM-4.6 model. (Exacto API example, model-variant documentation, GLM-4.6 catalog)

curl https://openrouter.ai/api/v1/responses \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -d '{
    "model": "z-ai/glm-4.6:exacto",
    "input": "Your prompt here"
  }'

Pricing depends on the serving route. OpenRouter’s GLM-4.6 page lists multiple providers and different input/output rates; those base-model listings should not be treated as one fixed Exacto price. Check the current catalog and the route used for your requests when estimating cost. Separately, Z.ai’s direct API pricing page lists GLM-4.6 in USD per million tokens at $0.60 per million input tokens and $2.20 per million output tokens. Those are Z.ai direct API rates for base GLM-4.6, not a quoted price for OpenRouter Exacto. (OpenRouter GLM-4.6 pricing, Z.ai pricing)

Before adoption, check the facts that affect your deployment:

1. Model and route: confirm z-ai/glm-4.6:exacto is accepted and inspect the serving route used.
2. Tool behavior: test your actual tools, including malformed arguments, tool errors, and multi-step calls.
3. Limits and compatibility: verify context/output limits and request parameters for the selected route.
4. Cost: compare route-specific prices and account for input, output, retries, and tool-call rounds.
5. Operations: measure latency and test failure handling and fallback behavior in your own deployment.

A practical selection test

Run base GLM-4.6 and the Exacto route on the same representative production tasks. Include normal tool calls and cases where tools return empty, unexpected, or erroneous results. Score task completion, correct tool choice, valid arguments, recovery from errors, latency, and total token cost. Keep prompts, tools, and evaluation criteria consistent.

Choose Exacto if the measured improvement in your tool-use outcomes justifies its route-specific cost and operational trade-offs. If it does not, default routing may be a better fit. OpenRouter positions Exacto for workflows that value tool-calling reliability over raw cost efficiency; validate that trade-off on your workload. (Exacto documentation)

Sources

Exacto Variant - Quality-first provider routing for tool use

Model Variants - Catalog and Routing Suffixes

GLM 4.6 - API Pricing & Benchmarks | OpenRouter

Migrate to GLM-4.6 - Overview - Z.AI Developer Document

Pricing - Overview - Z.AI Developer Document

GLM-4.6 vs Sonnet 4.5 — Open-Source Diff-Edit Convergence | Cline Report