{"apiVersion":"v1","taxonomyVersion":1,"generatedAt":"2026-10-29T20:43:05.137Z","data":[{"versionId":"ver_GAKou8jbMjL6","versionNumber":35,"effectiveAt":"2026-10-28T00:15:38.667Z","correctionState":"current","confidence":"high","score":0.981,"changeKind":"evidence_update","changedFields":["score"],"releaseNote":null},{"versionId":"ver_wwBfpeIV007B","versionNumber":34,"effectiveAt":"2026-10-26T01:24:55.342Z","correctionState":"current","confidence":"high","score":0.982,"changeKind":"evidence_update","changedFields":["score"],"releaseNote":null},{"versionId":"ver_elxJ0XyE_r2y","versionNumber":33,"effectiveAt":"2026-10-25T01:21:53.385Z","correctionState":"current","confidence":"high","score":0.983,"changeKind":"evidence_update","changedFields":["score"],"releaseNote":null},{"versionId":"ver_t9ZbFUnflxv1","versionNumber":32,"effectiveAt":"2026-10-24T01:52:51.311Z","correctionState":"current","confidence":"high","score":0.985,"changeKind":"evidence_update","changedFields":["score"],"releaseNote":null},{"versionId":"ver_J95mZ1rRWUD0","versionNumber":31,"effectiveAt":"2026-10-19T01:52:06.660Z","correctionState":"current","confidence":"high","score":0.987,"changeKind":"evidence_update","changedFields":["score"],"releaseNote":null},{"versionId":"ver_rvFuwACJM15U","versionNumber":30,"effectiveAt":"2026-10-18T02:42:04.549Z","correctionState":"current","confidence":"high","score":0.98,"changeKind":"material_change","changedFields":["capability","access","availability"],"releaseNote":null},{"versionId":"ver_aJ26u0UcrXi2","versionNumber":29,"effectiveAt":"2026-10-17T16:50:45.923Z","correctionState":"current","confidence":"high","score":0.98,"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_z5xUrMlgLVne","versionNumber":28,"effectiveAt":"2026-10-17T03:41:02.608Z","correctionState":"current","confidence":"high","score":0.978,"changeKind":"material_change","changedFields":["availability"],"releaseNote":null},{"versionId":"ver_rg9ude1s0gvW","versionNumber":27,"effectiveAt":"2026-10-16T16:41:48.295Z","correctionState":"current","confidence":"high","score":0.978,"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_y4XXn3-uwQfE","versionNumber":26,"effectiveAt":"2026-10-12T08:11:56.177Z","correctionState":"current","confidence":"high","score":0.977,"changeKind":"material_change","changedFields":["availability"],"releaseNote":null},{"versionId":"ver_4g1wDjZzTN5Q","versionNumber":25,"effectiveAt":"2026-10-11T22:00:42.668Z","correctionState":"current","confidence":"high","score":0.977,"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_iO1L7tkNN_VS","versionNumber":24,"effectiveAt":"2026-10-09T10:52:51.253Z","correctionState":"current","confidence":"high","score":0.975,"changeKind":"material_change","changedFields":["access","availability"],"releaseNote":null},{"versionId":"ver_lcLZ5rMSnlpQ","versionNumber":23,"effectiveAt":"2026-10-08T15:33:24.338Z","correctionState":"current","confidence":"high","score":0.975,"changeKind":"major_release","changedFields":["access","deployment","availability","release"],"releaseNote":"The b11507 pre-release adds support for MoE cache over multiple GPUs."},{"versionId":"ver_w9QKQ-WJnBZu","versionNumber":22,"effectiveAt":"2026-10-07T12:45:58.876Z","correctionState":"current","confidence":"high","score":0.986,"changeKind":"material_change","changedFields":["modality","access","deployment","availability"],"releaseNote":null},{"versionId":"ver_ycVkWQJ6NdwY","versionNumber":21,"effectiveAt":"2026-10-06T13:44:58.185Z","correctionState":"current","confidence":"high","score":0.972,"changeKind":"material_change","changedFields":["capability","modality","deployment","pricing","availability"],"releaseNote":null},{"versionId":"ver_trAgRae9zUhX","versionNumber":20,"effectiveAt":"2026-10-05T14:41:42.353Z","correctionState":"current","confidence":"high","score":0.986,"changeKind":"material_change","changedFields":["access","pricing","availability"],"releaseNote":null},{"versionId":"ver_v5LsVCwte-uq","versionNumber":19,"effectiveAt":"2026-10-05T09:04:23.154Z","correctionState":"current","confidence":"high","score":0.986,"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_nbeD5ZRbkqKM","versionNumber":18,"effectiveAt":"2026-10-05T03:02:22.942Z","correctionState":"current","confidence":"high","score":0.985,"changeKind":"major_release","changedFields":["access","pricing","deployment","availability","release"],"releaseNote":"Adds support for mixing embeddings and raw tokens in a batch."},{"versionId":"ver_Pe_S7Nm8gCJY","versionNumber":17,"effectiveAt":"2026-10-04T15:36:40.983Z","correctionState":"current","confidence":"high","score":0.983,"changeKind":"material_change","changedFields":["capability","modality","availability"],"releaseNote":null},{"versionId":"ver_kNemqiUysnVM","versionNumber":16,"effectiveAt":"2026-10-03T16:33:46.111Z","correctionState":"current","confidence":"high","score":0.966,"changeKind":"material_change","changedFields":["capability","pricing","availability","deployment"],"releaseNote":null},{"versionId":"ver_RlESaiIC8wnC","versionNumber":15,"effectiveAt":"2026-10-03T08:48:17.749Z","correctionState":"current","confidence":"high","score":0.966,"changeKind":"major_release","changedFields":["deployment","availability","release"],"releaseNote":"Adds initial text-only support for the Clef decision model."},{"versionId":"ver_lYuw9P-Z505h","versionNumber":14,"effectiveAt":"2026-10-03T02:46:17.667Z","correctionState":"current","confidence":"high","score":0.962,"changeKind":"major_release","changedFields":["pricing","capability","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_9gK_Pp-deZ3K","versionNumber":13,"effectiveAt":"2026-10-02T17:29:36.680Z","correctionState":"current","confidence":"high","score":0.956,"changeKind":"material_change","changedFields":["pricing","availability","deployment"],"releaseNote":null},{"versionId":"ver_aO2LU8yxJbRv","versionNumber":12,"effectiveAt":"2026-10-02T14:42:20.913Z","correctionState":"current","confidence":"high","score":0.956,"changeKind":"major_release","changedFields":["pricing","deployment","availability","release"],"releaseNote":"Release b11345 adds support for q2_k and q3_k quantization types on Hexagon."},{"versionId":"ver_GJwGfP-U9LZv","versionNumber":11,"effectiveAt":"2026-10-01T18:22:38.052Z","correctionState":"current","confidence":"high","score":0.948,"changeKind":"material_change","changedFields":["modality","access","deployment","pricing","availability","deployment"],"releaseNote":null},{"versionId":"ver_vk4EqyYf_kMw","versionNumber":10,"effectiveAt":"2026-09-30T20:28:32.972Z","correctionState":"current","confidence":"high","score":0.973,"changeKind":"major_release","changedFields":["capability","deployment","availability","release"],"releaseNote":"Adds support for dflash conversion and feature extraction for MiMo."},{"versionId":"ver_CSK6Ulsnkkiq","versionNumber":9,"effectiveAt":"2026-09-29T20:10:47.135Z","correctionState":"current","confidence":"high","score":0.97,"changeKind":"material_change","changedFields":["capability","modality","access","deployment","availability","deployment"],"releaseNote":null},{"versionId":"ver_s11ivKaJwx_D","versionNumber":8,"effectiveAt":"2026-09-29T02:14:25.262Z","correctionState":"current","confidence":"high","score":0.97,"changeKind":"major_release","changedFields":["access","modality","capability","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_znll7y19Asap","versionNumber":7,"effectiveAt":"2026-09-28T21:00:45.863Z","correctionState":"current","confidence":"high","score":0.959,"changeKind":"material_change","changedFields":["capability","modality","access","deployment","pricing","availability","deployment"],"releaseNote":null},{"versionId":"ver_IXs-m7cxjqNz","versionNumber":6,"effectiveAt":"2026-09-28T02:07:22.962Z","correctionState":"current","confidence":"high","score":0.922,"changeKind":"major_release","changedFields":["pricing","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_hsPXE5FV0MTX","versionNumber":5,"effectiveAt":"2026-09-26T22:49:43.632Z","correctionState":"current","confidence":"high","score":0.9,"changeKind":"material_change","changedFields":["access","deployment","availability","deployment"],"releaseNote":null},{"versionId":"ver_ze0SEeXKhypJ","versionNumber":4,"effectiveAt":"2026-09-25T23:41:42.537Z","correctionState":"current","confidence":"high","score":0.95,"changeKind":"material_change","changedFields":["capability","modality","access","deployment","pricing","availability","pricingModel","deployment","privacy"],"releaseNote":null},{"versionId":"ver_2UH99lb-a_jR","versionNumber":3,"effectiveAt":"2026-09-25T13:50:25.568Z","correctionState":"current","confidence":"high","score":0.95,"changeKind":"major_release","changedFields":["deployment","availability","release"],"releaseNote":"Adds support for sparse FA on SYCL."},{"versionId":"ver_BgtVSU38xI7q","versionNumber":2,"effectiveAt":"2026-09-24T19:44:31.298Z","correctionState":"current","confidence":"high","score":1,"changeKind":"major_release","changedFields":["pricing","capability","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_A6UiAuwoga3N","versionNumber":1,"effectiveAt":"2026-09-24T01:33:32.426Z","correctionState":"current","confidence":"high","score":1,"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."}}}