{"version":1,"type":"story","url":"https://digestai.news/story/continuity-research-releases-tess-4-5-27b-open-weights-coding-model","json":"https://digestai.news/story/continuity-research-releases-tess-4-5-27b-open-weights-coding-model.json","markdown":"https://digestai.news/story/continuity-research-releases-tess-4-5-27b-open-weights-coding-model.md","slug":"continuity-research-releases-tess-4-5-27b-open-weights-coding-model","headline":"Continuity Research releases Tess-4.5-27B open-weights coding model","summary":"Continuity Research released Tess-4.5, a 27-billion-parameter open-weights model designed for engineering and terminal work. Built by Migel Tissera as a fine-tune of Qwen3.8-27B, the model is available under an Apache 2.0 license and runs locally on consumer hardware, including single GPUs and laptops.\n\nAccording to Continuity Research, Tess-4.5 matches its base model on the BenchLocal evaluation suite, passing 128 of 150 scenarios on the first try and 133 within three attempts. The release notes state that the model generated 25% fewer tokens than Qwen3.8-27B across all 150 scenarios, including 32% fewer on the CLI-40 pack and 43% fewer on data extraction tasks.\n\nTess-4.5 includes a multi-token-prediction draft head for speculative decoding. Continuity Research reports that code generation speed increased from 49 to 96 tokens per second on an Nvidia RTX 4090, and from 24 to 38 tokens per second on an Apple MacBook Pro. Quantized GGUF files and vLLM deployment setups were released alongside the base weights.","keyPoints":["Tess-4.5 is a 27-billion-parameter open-weights fine-tune of Qwen3.8-27B released under the Apache 2.0 license","On BenchLocal, it passed 128 of 150 scenarios on the first try while using 25% fewer tokens than its base model","A multi-token-prediction draft head increased code generation from 49 to 96 tokens per second on an RTX 4090"],"whyItMatters":"Tess-4.5 demonstrates how fine-tuning can reduce token usage and improve inference speed on local hardware without sacrificing coding and tool-use performance.","category":{"slug":"models","name":"Generative AI & Models","url":"https://digestai.news/category/models"},"entities":{"companies":["Continuity Research","Thinking Machines Lab"],"models":["Tess-4.5","Tess-4.5-27B","Qwen3.8-27B"],"people":["Migel Tissera"]},"firstPublishedAt":"2026-10-20T00:09:50Z","updatedAt":"2026-10-20T00:09:50Z","sourceCount":1,"hasPrimarySource":true,"sources":[{"outlet":"huggingface.co","title":"migtissera/Tess-4.5-27B (Qwen finetune)","url":"https://huggingface.co/migtissera/Tess-4.5-27B","publishedAt":"2026-10-20T00:09:50Z","type":"primary","primary":true,"lead":true}],"sourceNotes":null,"discussions":[{"site":"Reddit","url":"https://www.reddit.com/r/LocalLLaMA/comments/1xadfm7/migtisseratess4527b_qwen_finetune/","points":null}],"thread":null,"cite":{"text":"Digest AI, \"Continuity Research releases Tess-4.5-27B open-weights coding model\", 20 October 2026, https://digestai.news/story/continuity-research-releases-tess-4-5-27b-open-weights-coding-model","publisher":"Digest AI","title":"Continuity Research releases Tess-4.5-27B open-weights coding model","datePublished":"2026-10-20T00:09:50Z","url":"https://digestai.news/story/continuity-research-releases-tess-4-5-27b-open-weights-coding-model"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}