Compare dated releases from OpenAI, DeepSeek, Google and Anthropic. Launches are verifiable events; vendor benchmark claims are not an independent ranking.
Selected releases through September 10, 2026; not an exhaustive model catalog or a best-model ranking.
Dates without a time use the local date of the event.
Google introduced Gemini 2.0 and made experimental Flash available to developers with multimodal input and text output.
Image and audio output were limited to early access; this was not general availability.
DeepSeek released R1 and made its reasoning model available through an API and downloadable model releases.
Distribution choices matter alongside capability: an API service and downloadable weights support different ways to build products.
Google introduced Gemini 2.5 with an experimental Pro model available through Google AI Studio and Gemini Advanced.
Experimental availability is a different milestone from a later stable release. The original launch post identifies which version was available.
Anthropic introduced Claude Opus 4 and Sonnet 4, offering both direct responses and extended thinking.
The release emphasized coding and longer agent workflows. Those are the vendor's product priorities, not proof of a universal winner.
DeepSeek's dated changelog records the release of V4.1 Flash with native visual input support through its API.
This extends the timeline beyond the initial reasoning-model launches. Published vendor scores remain separate from independent evaluation.
Selected releases through September 10, 2026; not an exhaustive model catalog or a best-model ranking.
Explore this branchAI chips: US export controls affecting ChinaExplore this branchEU AI Act: phased rules and the 2026 changes