English text is being prepared.

Hugging Face releases Phonon, an on-device speech model

On November 5, 2026, Hugging Face released Phonon, an on-device TTS model, offering two sizes of approximately 65 MB and 75 MB, covering English, French, German, Spanish, and Portuguese, with more than 100 built-in voices. Phonon 31m supports only English, is open-sourced under Apache 2.0, and can be used in commercial products; Phonon 40m covers five languages and uses a non-commercial license. The company says Phonon ranks first in German and Spanish intelligibility on the Open TTS Leaderboard and leads in CPU streaming first-packet latency benchmarks, and that its accompanying inference framework ptts supports more than 10 backends including CPU, GPU, and NPU. On the same day, Unite.AI reported that Gradium issued a technical follow-up statement saying Phonon surpassed 200,000 downloads within two weeks of launch, and provided 95th-percentile CPU first-audio latency figures: Phonon 31m English at 28 milliseconds, Phonon 40m English at 38 milliseconds, and Pocket TTS at 165 milliseconds; the report also said the open weights were released on October 20.

Hugging Face releases on-device speech model Phonon

On November 5, 2026, Hugging Face released Phonon, an on-device TTS model, offering two sizes of approximately 65 MB and 75 MB, covering English, French, German, Spanish, and Portuguese, with more than 100 built-in voices. Phonon 31m supports only English, is open-sourced under Apache 2.0, and can be used in commercial products; Phonon 40m covers five languages and uses a non-commercial license. The company says Phonon ranks first in German and Spanish intelligibility on the Open TTS Leaderboard and leads in CPU streaming first-packet latency benchmarks, and that its accompanying inference framework ptts supports more than 10 backends including CPU, GPU, and NPU. On the same day, Unite.AI reported that Gradium issued a technical follow-up statement saying Phonon surpassed 200,000 downloads within two weeks of launch, and provided 95th-percentile CPU first-audio latency figures: Phonon 31m English at 28 milliseconds, Phonon 40m English at 38 milliseconds, and Pocket TTS at 165 milliseconds; the report also said the open weights were released on October 20.

View story:Hugging Face releases Phonon, an on-device speech model

Sentient OS launches on-device AI that reads user data at night

On October 28, 2026, Sentient OS released on-device AI that at night uses a local LLM to read the user's messages, files, and notes, and writes what it learns into a pure markdown folder on the computer, which the user can open, edit, or delete. On the same day, Testing Catalog reported that Sentient released the open-source AI model Sentient OS, which can gather context from local files and messages and execute tasks on the user's computer; its Sidekick feature supports handing off tasks such as forms or emails open on the screen for it to complete using its own cursor. The report also said that what it learns is stored locally as plain text files, which users can open and edit; the first 1,000 users who enter FINAL_1000 in the app can use it free for life. Another report from the same day mentioned that during the day, Double Tap and Sidekick read these files to draft replies and complete tasks in apps, and only heavier tasks call the user-selected cloud model. There have been no further developments so far.

Sentient OS launches on-device AI that reads user data at night

On October 28, 2026, Sentient OS released on-device AI that at night uses a local LLM to read the user's messages, files, and notes, and writes what it learns into a pure markdown folder on the computer, which the user can open, edit, or delete. On the same day, Testing Catalog reported that Sentient released the open-source AI model Sentient OS, which can gather context from local files and messages and execute tasks on the user's computer; its Sidekick feature supports handing off tasks such as forms or emails open on the screen for it to complete using its own cursor. The report also said that what it learns is stored locally as plain text files, which users can open and edit; the first 1,000 users who enter FINAL_1000 in the app can use it free for life. Another report from the same day mentioned that during the day, Double Tap and Sidekick read these files to draft replies and complete tasks in apps, and only heavier tasks call the user-selected cloud model. There have been no further developments so far.

View story:Sentient OS launches on-device AI that reads user data at night

NVIDIA releases on-device small model Nemotron-3.5-Bolt-E4B

NVIDIA releases the on-device small model Nemotron-3.5-Bolt-E4B. A report on October 23, 2026 states that the model can run locally on devices with 6GB of memory, targeting on-device devices such as phones; it is trained based on Gemma 4, supports vision, and Unsloth simultaneously provides the GGUF file and a guide link. NVIDIA says it targets scenarios such as local assistants, interactive characters, and robots that can converse and call tools. A follow-up report on October 24 states that the model has 8B total parameters, 4B effective parameters, and a 64K context, and is post-trained using only reinforcement learning based on Google Gemma 4 E4B; the Q4_K_M version occupies 3.1 to 4.3 GB of VRAM on RTX and can run on GeForce RTX, Jetson, and DGX Spark. The report also states that in NVIDIA's own testing, it leads Gemma 4 E4B, Qwen3.5-4B, and Nemotron 3 Nano 4B in 15 of 21 benchmarks, but Tau2-Bench is 55.1, trailing Qwen3.5-4B's 75.1.

NVIDIA releases on-device small model Nemotron-3.5-Bolt-E4B

NVIDIA releases the on-device small model Nemotron-3.5-Bolt-E4B. A report on October 23, 2026 states that the model can run locally on devices with 6GB of memory, targeting on-device devices such as phones; it is trained based on Gemma 4, supports vision, and Unsloth simultaneously provides the GGUF file and a guide link. NVIDIA says it targets scenarios such as local assistants, interactive characters, and robots that can converse and call tools. A follow-up report on October 24 states that the model has 8B total parameters, 4B effective parameters, and a 64K context, and is post-trained using only reinforcement learning based on Google Gemma 4 E4B; the Q4_K_M version occupies 3.1 to 4.3 GB of VRAM on RTX and can run on GeForce RTX, Jetson, and DGX Spark. The report also states that in NVIDIA's own testing, it leads Gemma 4 E4B, Qwen3.5-4B, and Nemotron 3 Nano 4B in 15 of 21 benchmarks, but Tau2-Bench is 55.1, trailing Qwen3.5-4B's 75.1.

View story:NVIDIA releases on-device small model Nemotron-3.5-Bolt-E4B

Meta distills a 30B on-device agent model, Muse Glimmer

Meta distilled the dense 30B model Muse Glimmer from the larger Muse Spark, training it on synthetic agent trajectories in simulated applications. Its 4-bit weights are under 20 GB, allowing it to run on a single RTX 5090. According to a report dated October 22, 2026, on the multi-tool benchmark MCPAtlas, Muse Glimmer scored 75.5, higher than Gemma4-31B's 54.2 and Qwen3.6-27B's 62.5; however, on OSWorld-Verified it lagged behind Qwen3.6-27B's 75.6 with a score of 65.9. There are currently no subsequent developments.