Enum VisionModel 

pub enum VisionModel {
    Claude,
    Gpt4o,
    Gpt5,
    Gemini15,
    LlamaVision,
    QwenVl,
    DeepseekVl,
}

Supported vision model families with their patch pricing.

Variants

Claude

Claude 3.5/4.5/4.6/4.7: Area-based calculation (Tokens ≈ width × height / 750).

Gpt4o

GPT-4o / GPT-4.5 high detail: fits in 2048x2048, scales short side to 768, then 512x512 tiles.

Gpt5

GPT-5/5.5: 6000px max dim, 10.24M max pixels, 512×512 tiles, 1536 token cap.

Gemini15

Gemini 2.0/3.0: flat 258 tokens if ≤ 384x384, else 258 per 768x768 tile.

LlamaVision

Meta Llama 3.2/3.3 Vision (Mllama): 560×560 tiles, aspect-ratio canvas capped at 4 tiles. (Llama 4 uses a different native-multimodal vision encoder and is not modeled by this arm.)

QwenVl

Alibaba Qwen2-VL / Qwen2.5-VL: 28px effective grid (14px ViT patch × 2×2 merge), token clamp.

DeepseekVl

DeepSeek-VL2: 384×384 base + dynamic 384 local tiles (open-weights; value is local-context savings).

Trait Implementations

impl Clone for VisionModel

fn clone(&self) -> VisionModel

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Copy for VisionModel

impl Debug for VisionModel

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.