// Model routing matrix

Model catalogue

Prices are per million tokens in US dollars. Volume customers receive deeper tiered discounts.

Claude Opus 4

Anthropic

−65% save

Deepest reasoning, long analysis and agentic work.

Context
200K
Input
$15.00$5.25 / 1M
Output
$75.00$26.25 / 1M
claude-opus-4

Claude Sonnet 4

Anthropic

−65% save

The everyday workhorse for coding and document work.

Context
200K
Input
$3.00$1.05 / 1M
Output
$15.00$5.25 / 1M
claude-sonnet-4

Claude Haiku 4

Anthropic

−65% save

Fast and cheap for classification and high-volume calls.

Context
200K
Input
$0.80$0.28 / 1M
Output
$4.00$1.40 / 1M
claude-haiku-4

GPT-5

OpenAI

−30% save

Strongest general-purpose model with broad tool support.

Context
400K
Input
$1.25$0.88 / 1M
Output
$10.00$7.00 / 1M
gpt-5

GPT-4.1

OpenAI

−30% save

Long-context coding and refactoring at enterprise scale.

Context
1M
Input
$2.00$1.40 / 1M
Output
$8.00$5.60 / 1M
gpt-4.1

o3

OpenAI

−30% save

Step-by-step reasoning for maths, planning and audits.

Context
200K
Input
$2.00$1.40 / 1M
Output
$8.00$5.60 / 1M
o3

Codex mini

OpenAI

−55% save

Dedicated discount channel for coding agents.

Context
200K
Input
$1.50$0.68 / 1M
Output
$6.00$2.70 / 1M
codex-mini

Gemini 3.1 Pro

Google

−40% save

Two million tokens of context across text, image, audio and video.

Context
2M
Input
$2.50$1.50 / 1M
Output
$15.00$9.00 / 1M
gemini-3-1-pro

Gemini 2.5 Flash

Google

−40% save

Cheap, fast multimodal for bulk document and media pipelines.

Context
1M
Input
$0.30$0.18 / 1M
Output
$2.50$1.50 / 1M
gemini-2-5-flash

Gemini 3.1 Flash Image

Google

−15% save

Native image generation and editing, priced per image.

Context
—
Input
$0.04$0.03 / image
gemini-3-1-flash-image

Seedance 2.5

ByteDance

−30% save

Scarce video generation capacity — full version, no queue.

Context
—
Input
$0.60$0.42 / clip
seedance-2-5

Wan 3.0

Alibaba

−30% save

Cinematic video generation for studios, agencies and MCNs.

Context
—
Input
$0.50$0.35 / clip
wan-3-0

Video and image generation capacity is limited. Reserve ahead for production workloads.