视觉 API 适合截图、收据、商品图、工单附件。关键是模型能不能吃你的图、要不要 JSON 输出、在你自己的样本上错得有多离谱。价格放第二。这页只放一小撮候选，方便你 A/B，而不是扫整本目录。
来源依据: NextModel 的能力映射，以及可用时的 OpenRouter 输入模态元数据。 · 更新日期 2026-07-01
步骤
匹配分
先从候选名单开始，再用真实提示词测试，并在接入生产路由前比较月度成本。
Claude Haiku 4.5 is Anthropic’s fastest and most efficient model, delivering near-frontier intelligence at a fraction of the cost and latency of larger Claude models. Matching Claude Sonnet 4’s performance...
Claude Opus 4.5 is Anthropic’s frontier reasoning model optimized for complex software engineering, agentic workflows, and long-horizon computer use. It offers strong multimodal capabilities, competitive performance across real-world coding and...
Opus 4.6 is Anthropic’s strongest model for coding and long-running professional tasks. It is built for agents that operate across entire workflows rather than single prompts, making it especially effective...
Claude Sonnet 4.5 is Anthropic’s most advanced Sonnet model to date, optimized for real-world agents and coding workflows. It delivers state-of-the-art performance on coding benchmarks such as SWE-bench Verified, with...
比较表
缩小生产候选、建立兜底策略或比较模型经济性时用。
| Model | Provider | Input | Output | Context | Capabilities | Best for | Latency | Status | Source |
|---|---|---|---|---|---|---|---|---|---|
| Anthropic: Claude Haiku 4.5anthropic/claude-haiku-4.5 | Anthropic | $1 / 1M tokens | $5 / 1M tokens | 200k | StreamingTool callingJSON modeVision | image understanding, multimodal chat | 1000-3000ms | Catalog | Platform curated |
| Anthropic: Claude Opus 4.5anthropic/claude-opus-4.5 | Anthropic | $5 / 1M tokens | $25 / 1M tokens | 200k | StreamingTool callingJSON modeVision | image understanding, multimodal chat | 1000-3000ms | Catalog | Platform curated |
| Anthropic: Claude Opus 4.6anthropic/claude-opus-4.6 | Anthropic | $5 / 1M tokens | $25 / 1M tokens | 1M | StreamingTool callingJSON modeVision | image understanding, multimodal chat | 1000-3000ms | Catalog | Platform curated |
| Anthropic: Claude Sonnet 4.5anthropic/claude-sonnet-4.5 | Anthropic | $3 / 1M tokens | $15 / 1M tokens | 1M | StreamingTool callingJSON modeVision | image understanding, multimodal chat | 1000-3000ms | Catalog | Platform curated |
| Anthropic: Claude Sonnet 4.6anthropic/claude-sonnet-4.6 | Anthropic | $3 / 1M tokens | $15 / 1M tokens | 1M | StreamingTool callingJSON modeVision | image understanding, multimodal chat | 1000-3000ms | Catalog | Platform curated |
| FLUX.2 Flexblack-forest-labs/FLUX.2-flex | Black Forest Labs | $0.2 / image | — | — | StreamingVision | image understanding, multimodal chat | 1000-3000ms | Catalog | Platform curated |
| FLUX.2 PROblack-forest-labs/FLUX.2-pro | Black Forest Labs | $0.075 / image | — | — | StreamingVision | image understanding, multimodal chat | 1000-3000ms | Catalog | Platform curated |
| Doubao Seedance 2 0 260128doubao/doubao-seedance-2-0-260128 | Doubao | $0.148 / s | — | — | StreamingVision | image understanding, multimodal chat | 1000-3000ms | Catalog | Platform curated |
常见问题
图像输入支持、是否需要 JSON、延迟、输出成本，以及真实图片上的质量。合成基准经常骗人。
轻量任务可以。密文档和高精度抽取通常需要更强模型和真实评测集。
相关
指南