{"apiVersion":"v1","taxonomyVersion":1,"generatedAt":"2026-10-30T15:05:29.686Z","data":[{"versionId":"ver_qK9lAYzyF4GG","versionNumber":21,"effectiveAt":"2026-10-30T00:39:46.731Z","correctionState":"current","confidence":"high","score":0.959,"changeKind":"evidence_update","changedFields":["score"],"releaseNote":null},{"versionId":"ver_tE9fH_EYlpQj","versionNumber":20,"effectiveAt":"2026-10-26T01:00:55.301Z","correctionState":"current","confidence":"high","score":0.962,"changeKind":"evidence_update","changedFields":["score"],"releaseNote":null},{"versionId":"ver_NyeS0zErSE4y","versionNumber":19,"effectiveAt":"2026-10-25T00:58:53.376Z","correctionState":"current","confidence":"high","score":0.963,"changeKind":"evidence_update","changedFields":["score"],"releaseNote":null},{"versionId":"ver_xEV7ZPgw7bbE","versionNumber":18,"effectiveAt":"2026-10-24T01:29:51.296Z","correctionState":"current","confidence":"high","score":0.964,"changeKind":"evidence_update","changedFields":["score"],"releaseNote":null},{"versionId":"ver_AtcAmoWEl3QQ","versionNumber":17,"effectiveAt":"2026-10-17T16:50:45.923Z","correctionState":"current","confidence":"high","score":0.97,"changeKind":"major_release","changedFields":["access","capability","targetUser","availability","release"],"releaseNote":"Pre-release b11714 adds a Discover Models view for browsing and downloading GGUF models from Hugging Face, with search, model details, download tracking, and pause/resume support."},{"versionId":"ver_G0esCEYTD4i4","versionNumber":16,"effectiveAt":"2026-10-16T16:41:48.295Z","correctionState":"current","confidence":"high","score":0.967,"changeKind":"major_release","changedFields":["deployment","availability","release"],"releaseNote":"Release b11695 adds MTP support for Mellum and Mellum2.1, along with inline RoPE construction for Mellum."},{"versionId":"ver_eXX8QWRqEOCB","versionNumber":15,"effectiveAt":"2026-10-11T22:00:42.668Z","correctionState":"current","confidence":"high","score":0.964,"changeKind":"major_release","changedFields":["deployment","availability","release"],"releaseNote":"This pre-release adds OpenCL GPU decode support for Gemma-4 E4B and E2B, alongside OpenCL improvements and optimizations."},{"versionId":"ver_q31uVngGj5HM","versionNumber":14,"effectiveAt":"2026-10-08T15:33:24.338Z","correctionState":"current","confidence":"high","score":0.961,"changeKind":"major_release","changedFields":["deployment","availability","release"],"releaseNote":"The b11507 pre-release adds support for MoE cache over multiple GPUs."},{"versionId":"ver_TjrwiF4HaxJK","versionNumber":13,"effectiveAt":"2026-10-05T09:04:23.154Z","correctionState":"current","confidence":"high","score":0.972,"changeKind":"major_release","changedFields":["deployment","availability","release"],"releaseNote":"Release b11408 adds an accelerated Q8_0 matrix kernel for the SpacemiT X60, enabling an accelerated path for that quantization on the device."},{"versionId":"ver_6I_gQnCmdwdC","versionNumber":12,"effectiveAt":"2026-10-05T03:02:22.942Z","correctionState":"current","confidence":"high","score":0.969,"changeKind":"major_release","changedFields":["deployment","availability","release"],"releaseNote":"Adds support for mixing embeddings and raw tokens in a batch."},{"versionId":"ver_NGFf7zDkRPK2","versionNumber":11,"effectiveAt":"2026-10-03T16:33:46.111Z","correctionState":"current","confidence":"high","score":0.965,"changeKind":"evidence_update","changedFields":["score"],"releaseNote":null},{"versionId":"ver_o5BiyXdsZhkS","versionNumber":10,"effectiveAt":"2026-10-03T08:48:17.749Z","correctionState":"current","confidence":"high","score":0.948,"changeKind":"major_release","changedFields":["deployment","availability","release"],"releaseNote":"Adds initial text-only support for the Clef decision model."},{"versionId":"ver_eyYknuuPMe5v","versionNumber":9,"effectiveAt":"2026-10-03T02:46:17.667Z","correctionState":"current","confidence":"high","score":0.942,"changeKind":"major_release","changedFields":["deployment","availability","release"],"releaseNote":"Release b11368 adds probabilistic sampling for simple draft and MTP speculative decoding, including rejection sampling support for grammar-constrained requests; probabilistic draft sampling is controlled by a flag and defaults to greedy."},{"versionId":"ver_PgF8Ctxeq5IU","versionNumber":8,"effectiveAt":"2026-10-02T17:29:36.680Z","correctionState":"current","confidence":"high","score":0.932,"changeKind":"evidence_update","changedFields":["score"],"releaseNote":null},{"versionId":"ver_tzZr1AXINLc8","versionNumber":7,"effectiveAt":"2026-10-02T14:42:20.913Z","correctionState":"current","confidence":"high","score":0.955,"changeKind":"major_release","changedFields":["deployment","availability","release"],"releaseNote":"Release b11345 adds support for q2_k and q3_k quantization types on Hexagon."},{"versionId":"ver_IIpHH-S2CQk9","versionNumber":6,"effectiveAt":"2026-09-30T20:28:32.972Z","correctionState":"current","confidence":"high","score":0.946,"changeKind":"major_release","changedFields":["deployment","availability","release"],"releaseNote":"Adds support for dflash conversion and feature extraction for MiMo."},{"versionId":"ver_-2OxHerSrRkx","versionNumber":5,"effectiveAt":"2026-09-29T02:14:25.262Z","correctionState":"current","confidence":"high","score":0.939,"changeKind":"major_release","changedFields":["modality","deployment","availability","release"],"releaseNote":"Release b11240 adds typed multimodal content support for vision, audio, and video inputs to the server’s /v1/embeddings endpoint, including support for Qwen3-VL-Embedding."},{"versionId":"ver_NwzoCARUg02W","versionNumber":4,"effectiveAt":"2026-09-28T02:07:22.962Z","correctionState":"current","confidence":"high","score":0.918,"changeKind":"major_release","changedFields":["access","modality","capability","deployment","availability","release"],"releaseNote":"The server can split RANK-pooling inputs for causal rerankers, enabling chunked prefill for long-document and multimodal reranking."},{"versionId":"ver_2h2hUEZKaV1C","versionNumber":3,"effectiveAt":"2026-09-25T13:50:25.568Z","correctionState":"current","confidence":"high","score":0.892,"changeKind":"major_release","changedFields":["deployment","availability","release"],"releaseNote":"Adds support for sparse FA on SYCL."},{"versionId":"ver_uxvlSzCDrgYy","versionNumber":2,"effectiveAt":"2026-09-24T19:44:31.298Z","correctionState":"current","confidence":"high","score":0.922,"changeKind":"major_release","changedFields":["pricing","deployment","availability","release"],"releaseNote":"Pre-release b11163 adds the llama_batch_ext API, including support for token, embedding, and state inputs in a batch."},{"versionId":"ver_lNZ4RVG90N3b","versionNumber":1,"effectiveAt":"2026-09-24T01:33:36.064Z","correctionState":"current","confidence":"high","score":0.86,"changeKind":"publication","changedFields":[],"releaseNote":null}],"nextCursor":null,"license":{"catalog":{"name":"CC BY 4.0","spdx":"CC-BY-4.0","url":"https://creativecommons.org/licenses/by/4.0/","attribution":"wowsuchai","attributionUrl":"https://wowsuchai.withglint.app","covers":["catalog metadata","classifications","generated summaries","feed event metadata"]},"excluded":{"covers":["source quotations","linked source content"],"terms":"Not licensed by wowsuchai. Source quotations and linked source content keep their original rights."}}}