Continuity Research releases Tess-4.5-27B open-weights coding model
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.
Key points
- 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
According 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.
Tess-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.
migtissera/Tess-4.5-27B (Qwen finetune)
huggingface.co · 20 October 2026
Loading the full article…
This text was published by huggingface.co. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
1source- Reddit discussionreddit.com
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Inception launches Mercury 3 diffusion LLM, claims parity with GPT-6 Luna and Claude Haiku 5.5 at 1,000+ tokens/sec · 3 src
- Nace AI releases Drex Max multimodal decision model for computer-use agents · 1 src
- Tel Aviv-based doubleAI releases debut code execution model Maxwell and raises $200M-$220M · 3 src
- Cloudflare launches Clairic extraction model on Workers AI preview · 2 src
- OpenAI says its model is better at COBOL, Fortran, and Visual Basic · 1 src
Comments
via GitHub Discussions