Which reasoning effort values work for SVG generation?

Short answer

svg.reasoning.effort controls how much reasoning the text model that authors the SVG spends, and which values are accepted depends on svg.model, not on the gateway:

Mainstream author models (gpt-5.6-terra, gpt-6-luna, and so on): none / low / medium / high / xhigh / max — minimal returns 400.
minimal belongs to the original GPT-5 generation (gpt-5 / gpt-5-mini / gpt-5-nano). Since gpt-5.1 OpenAI replaced it with none, and no later model accepts it.
Omitting reasoning, sending reasoning: null, or sending effort: null all generate normally and fall back to the author model’s default.

So “minimal is not supported” does not mean you typed the field wrong — that author model simply does not accept the tier. Changing the tier or the model fixes it.

With output_format: "svg", svg.model is required; leaving it out returns 400 SVG generation requires 'svg.model'. Tier questions only apply inside that object.

Keep the field nested inside `svg`

{
  "model": "gpt-image-2.6-sunburst",
  "prompt": "A minimal flat vector icon of a rocket, exactly three solid colors, geometric editable shapes, no text, no raster texture, transparent background.",
  "output_format": "svg",
  "svg": {
    "model": "gpt-5.6-terra",
    "reasoning": { "effort": "high" }
  },
  "quality": "low",
  "size": "1024x1024",
  "n": 1
}

Do not use a flat top-level field. Writing "svg_reasoning_effort": "high" raises no error and has no effect — the gateway drops the unknown field and treats the request as if no effort were sent. In our tests the flat form with minimal returned 200, while the same value under svg.reasoning.effort returned 400. A 200 alone does not prove the field took effect.

What each author model accepts

Measured on api.apiyi.com on 2026-11-10. The tier lists come from what each model echoes back in its own error message (probed on /v1/responses; the SVG path was verified end to end for gpt-5.6-terra, gpt-6-luna, gpt-5-nano, gpt-5 and gpt-4.1-mini):

`svg.model`	Supported tiers	`minimal`
`gpt-5.6-luna` / `gpt-5.6-sol` / `gpt-5.6-terra`	`none` / `low` / `medium` / `high` / `xhigh` / `max`	❌ 400
`gpt-6-luna` / `gpt-6-sol` / `gpt-6.1-terra`	`none` / `low` / `medium` / `high` / `xhigh` / `max`	❌ 400
`gpt-6-astra` / `gpt-6.1-sol`	`low` / `medium` / `high` / `xhigh` / `max` (no `none`)	❌ 400
`gpt-5.1` / `gpt-5.2` / `gpt-5.4` / `gpt-5.4-mini` / `gpt-5.5`	`none` / `low` / `medium` / `high` (`xhigh` from 5.2 on)	❌ 400
`gpt-5` / `gpt-5-mini` / `gpt-5-nano`	`minimal` / `low` / `medium` / `high`	✅ 200
`gpt-5-codex` / `gpt-5.1-codex` series	`low` / `medium` / `high`	❌ 400
`gpt-4.1` (non-reasoning, measured)	Does not accept `reasoning` at all	❌ 400

Unsure what a model supports? Send a value it probably rejects (max or minimal) and read the error — the upstream lists every tier that model accepts, which is faster and more reliable than guessing from docs.

Reading the errors

1. Tier not accepted by the author model (`minimal` is the usual case)

{
  "error": {
    "message": "Unsupported value: 'minimal' is not supported with the 'gpt-5.6-terra' model. Supported values are: 'none', 'low', 'medium', 'high', 'xhigh', and 'max'.",
    "localized_message": "Unknown error",
    "type": "shell_api_error",
    "param": "",
    "code": "unsupported_value"
  }
}

param is empty and the message names the model — this is validation from the author model, not the gateway. The request never reaches generation and is not billed.

2. Value outside the enum (typo, empty string)

{
  "error": {
    "message": "Provider API error: Invalid value: 'ultra'. Supported values are: 'none', 'minimal', 'low', 'medium', 'high', 'xhigh', and 'max'.",
    "localized_message": "Unknown error",
    "type": "shell_api_error",
    "param": "svg.reasoning.effort",
    "code": "invalid_value"
  }
}

param points at svg.reasoning.effort and seven values are listed — that is the gateway enum. It includes minimal, but that does not mean your author model supports it; error 1 is still the authority.

3. Author model has no reasoning at all

{
  "error": {
    "message": "Unsupported parameter: 'reasoning.effort' is not supported with this model.",
    "localized_message": "Unknown error",
    "type": "shell_api_error",
    "param": "",
    "code": "unsupported_parameter"
  }
}

This is what you get when a non-reasoning model such as gpt-4.1-mini authors the SVG — every tier fails, so drop the reasoning object entirely.

If you really want `minimal`

Switch the author model to the original GPT-5 family; SVG generation works in our tests:

{
  "model": "gpt-image-2.6-sunburst",
  "prompt": "A minimal flat vector icon of a single blue circle, two solid colors, geometric shapes, no text, transparent background.",
  "output_format": "svg",
  "svg": {
    "model": "gpt-5-nano",
    "reasoning": { "effort": "minimal" }
  },
  "quality": "low",
  "size": "1024x1024",
  "n": 1
}

It returns 200 with svg_usage.reasoning_tokens at 0 (the same prompt at low reports 320).

Do not downgrade svg.model to the original gpt-5 just to save a little authoring cost. minimal only trims the author model’s thinking tokens, which are already cheap, while the first-generation model is visibly weaker at vector code structure, path precision and prompt adherence than gpt-5.6-terra / gpt-6-luna — you may pay for it in rework. To cut cost, lower quality and size first; that is the larger, reference-image side of the bill.

How the tier affects output and cost

It only affects the author model, not the reference image. The image model renders the reference image first (billed under usage), then the text model thinks and writes the SVG (billed under svg_usage); both count.
reasoning_tokens is already included in output_tokens — do not add them twice. Billing uses the author model’s output price times svg_usage.output_tokens.
Higher tiers cost more. One max run produced 4,816 reasoning tokens and 4,987 total svg_usage.output_tokens; none is consistently 0.
But reasoning_tokens cannot prove a tier took effect. On the same simple prompt the low / medium / high / xhigh ranges overlap heavily (medium even reported 0 once); only none (always 0) and max (thousands) separate cleanly. low / medium is enough for simple shapes; go above high only for complex illustrations and multi-layer structure.

GPT Image 2.6 capabilities

SVG, GIF, complexity and markup-guided editing, with real results.

GPT-Image series overview

Specs, pricing, endpoints and common errors for all five models.

Why output tokens are so high

How resolution and quality drive output_tokens and per-image cost.

Model pricing

Live unit prices for author and image models — authoritative.

Basics

Basic API

Image API (Official)

Video API (Official)

Multimodal Understanding API

Text Generation

Audio API (Official)

Retrieval & Safety

Which reasoning effort values work for SVG generation?

Short answer

Keep the field nested inside `svg`

What each author model accepts

Reading the errors

1. Tier not accepted by the author model (`minimal` is the usual case)

2. Value outside the enum (typo, empty string)

3. Author model has no reasoning at all

If you really want `minimal`

How the tier affects output and cost

GPT Image 2.6 capabilities

GPT-Image series overview

Why output tokens are so high

Model pricing

​Short answer

​Keep the field nested inside svg

​What each author model accepts

​Reading the errors

​1. Tier not accepted by the author model (minimal is the usual case)

​2. Value outside the enum (typo, empty string)

​3. Author model has no reasoning at all

​If you really want minimal

​How the tier affects output and cost

​Related docs

GPT Image 2.6 capabilities

GPT-Image series overview

Why output tokens are so high

Model pricing

Short answer

Keep the field nested inside `svg`

What each author model accepts

Reading the errors

1. Tier not accepted by the author model (`minimal` is the usual case)

2. Value outside the enum (typo, empty string)

3. Author model has no reasoning at all

If you really want `minimal`

How the tier affects output and cost

Related docs