AI Model Launch Tracker

What we captured

NVIDIA and the University of Washington’s Institute for Protein Design announced RFD4-Proteina, an open protein-design model, and said it is available through named partners.

• • Combines programmable protein design with search capabilities.
• • Available through named partners including Amazon Bio Discovery, Anthropic’s Claude Science, Benchling, Databricks, Google Cloud, Hugging Face, Nebius, OpenAI, and Tamarind Bio.
• • NVIDIA reports higher computational design success rates and generation throughput than evaluated baselines across four benchmark suites in testing described in a research draft.

Source evidence

Confidence 96% / Completeness 87%

What we captured

NVIDIA published Kumo Forecast model weights and code for an Apache 2.0-licensed multivariate time-series forecasting model.

• • The article says Kumo Forecast is available under Apache 2.0.
• • The article identifies NVIDIA/Kumo-TS as the code repository and nvidia/Kumo-Forecast as the model-weights repository.

Source evidence

Confidence 95% / Completeness 91%

What we captured

Conway Research founder Sigil Wen released Bark 3.0, a 163 MB model built from NVIDIA's Parakeet, with weights available under CC-BY 4.0.

• • Released October 21, 2026, by Conway Research founder Sigil Wen.
• • The model is 163 MB and built from NVIDIA's Parakeet.
• • Weights are available under CC-BY 4.0.
• • The article says the published benchmark puts Bark 3.0 level with Parakeet on average.

Source evidence

Confidence 97% / Completeness 91%

What we captured

Gradium released the Phonon 31m and 40m family of on-device text-to-speech models, including a five-language multilingual model.

• • The family includes 31M- and 40M-parameter models.
• • The 40m multilingual model supports English, French, German, Spanish, and Portuguese.
• • Gradium describes the models as built for local inference without a network connection.

Source evidence

Confidence 94% / Completeness 91%

What we captured

ElevenLabs released two speech-to-speech models for real-time conversations, available through ElevenAgents and an OpenAI-compatible Realtime API.

• • The two models are named Eleven Realtime v1 Max and v1 Mini.
• • ElevenLabs reports that v1 Max scored 98.8% on Big Bench Audio with a 0.57-second response time.
• • Both models are available through ElevenAgents and an OpenAI-compatible Realtime API.

Source evidence

Confidence 99% / Completeness 87%

What we captured

GoTo announced a 27-billion-parameter model for Bahasa Indonesia and four regional languages, offering it free to use, study, and adapt.

• • Announced October 19, 2026, in Sanur, Bali.
• • Described as a 27-billion-parameter model.
• • Supports Bahasa Indonesia, Javanese, Sundanese, Balinese, and Bataknese.
• • Kontan reported the identifier gemma3-27b-cpt-gotoai-v3-instruct.
• • The article says the exact license was not disclosed.

Source evidence

Confidence 92% / Completeness 87%

What we captured

fal announced FLUX 3 Max, a video-generation model post-trained by fal Research in collaboration with Black Forest Labs, and made it available via fal's inference API.

• • The model is available for text-to-video, image-to-video, and keyframes-to-video generation through fal's API.
• • fal reports a 1,139 Elo score across 4,868 votes in preliminary Artificial Analysis Video Arena evaluations.
• • fal's benchmark reports 10.95 seconds to generate a 10-second video.

Source evidence

Confidence 99% / Completeness 87%

What we captured

MiniMax released M3.1 Flash, available through its API, M Plan subscription, and MiniMax Code desktop app.

• • Released October 19, 2026.
• • MiniMax claims 70% lower agent-task costs and nearly three times faster inference than its previous model.
• • Published prices are $0.15 per million input tokens and $0.60 per million output tokens.
• • MiniMax says model weights will be released later.

Source evidence

What we captured

fal launched FLUX 3 Max, developed with Black Forest Labs, with public text-to-video and image-to-video API endpoints.

• • fal announced the model on October 19, 2026.
• • fal reports 12.45 seconds to generate a 10-second clip, compared with 277.80 seconds for Seedance 2.5 in its published comparison.
• • Artificial Analysis lists FLUX 3 Max at 1,138 Elo in its October 19 image-to-video leaderboard snapshot.

Source evidence

Confidence 97% / Completeness 87%

What we captured

DeepSeek released V4.1-Flash-1017, an upgraded version of V4.1-Flash, with claimed improvements in speed and several capabilities.

• • DeepSeek released V4.1-Flash-1017 as an upgrade to V4.1-Flash.
• • DeepSeek listed output-token API prices of $0.59 per million off-peak and $1.19 per million at peak hours.
• • DeepSeek's own DeepSWE v1.1 evaluation gave the model a score of 77.0.

Source evidence

Confidence 96% / Completeness 79%