The distinction between first-party and integrated-third-party matters a lot here. Anthropic has historically focused on text and reasoning rather than multimodal generation. Their approach to images has been input-only (vision) not output.

But the competitive pressure is real: OpenAI has native image gen in GPT-4o, Google has Imagen in Gemini. Being the only major frontier lab without image output is increasingly a product gap, not a strategic choice.

The question is whether Anthropic builds vs partners. Building in-house image gen requires different training infrastructure, different safety evaluation frameworks, and different compute allocation. A partnership (like licensing from Stability or similar) would be faster but would not resolve this market YES.

45% feels reasonable. The product pressure pushes toward YES, but Anthropic's institutional culture and resource allocation push toward prioritizing reasoning capabilities over generation. I'd lean slightly below 45% — maybe 38-40%.

Does it have to be first party?

@Velaris very good question, i think so yes, right now claude code is able to call various image-generation apis but imo the friction is that it is not a native skill + users have to carry the cost which is not ideal. anthropic could still do deep integration for any open-source model and carry the cost which would solve this problem but not suffice to resolve this market yes

@Velaris @traders /Soli/will-anthropic-natively-integrate-i new related market with a more lenient resolution criteria

It seems so counterintuitive, why would they do that? It's a minefield rife with bad press and has nothing to do with their goals. They're one of the most popular AIs for companies, they don't need people using their models to generate memes.

@Velaris image models are very good at creating different kinds of visual assets and are critical for automating work. ~5-10% of my claude code usage involves it calling the nano-banana-pro api

Why is this so high?

Does a wrapper around another image model, e.g. nano banana count?

@creator Do you mean a dedicated standalone image model? Like Nano Banana? Or just regular Sonnet/Opus but which can output images?

@RS both would suffice