Best AI Models in Brazil - Page 22
View:
Compare the Top AI Models in Brazil as of July 2026 - Page 22
Sort By:
-
1
Aion 1.0 Instruct
Microsoft
Aion-1.0-Instruct is a pre-release small language model introduced in Microsoft Edge as a developer preview for early testing and feedback. It is designed to power Edge’s on-device Prompt and Writing Assistance APIs, giving web developers a faster, smaller, and more efficient model for AI-powered browser experiences. Microsoft previously used Phi-4-mini for these APIs, but its hardware requirements limited availability across devices. Aion-1.0-Instruct expands support to significantly more devices, including machines with less capable GPUs and, through CPU inference, devices without a GPU, while still delivering strong quality for a wide range of web use cases. The model is available in Edge Canary and Dev channels, allowing developers to evaluate it in real-world web scenarios, test API interoperability, and provide feedback before final optimizations. Aion-1.0-Instruct is meant to help developers build AI features directly into websites and browser extensions. -
2
Aion 1.0 Plan
Microsoft
Aion 1.0 Plan is Microsoft’s local agentic reasoning model for Windows, designed to bring fully agentic workflows onto the device without cloud dependency or per-token cost. It is a 14-billion-parameter reasoning and tool-calling model with a 32K context length, shipping in-box as part of Windows on capable devices. Unlike smaller on-device models focused on everyday text intelligence, Aion 1.0 Plan is built for local agentic reasoning, enabling applications to understand user intent, invoke tools, manage files, and orchestrate sub-agents directly on the device. It belongs to Microsoft’s new generation of on-device small language models purpose-built for local execution, representing the progression from efficient text intelligence at scale to more capable local planning and action. Aion 1.0 Plan is part of Windows’ broader push toward “unmetered intelligence,” where frontier models handle the hardest problems while local models support continuous, lower-cost agent workflows. -
3
Miso TTS
Miso TTS
Miso Labs builds emotive foundation models for voice, designed to help developers create voice agents that feel fast, warm, and human instead of robotic or delayed. Its flagship model, Miso TTS, is an 8-billion-parameter transformer model for state-of-the-art emotive speech and dialogue generation, with open source weights available on Hugging Face and API access coming soon. Miso is built for real-time conversational voice, responding in 110ms to preserve natural flow and avoid the awkward pauses common in AI voice agents. It supports one-shot voice cloning, allowing users to clone a voice from a ten-second audio clip while keeping the agent’s voice consistent from the first second of a call to the last. Miso Labs also emphasizes local and sovereign deployment, with open source models built for local use and on-premises hosting and support available for enterprise teams that need to keep sensitive data in-house. -
4
Holo3.1
H Company
Holo3.1 is H Company’s family of fast and local computer-use agents, built to operate across web, desktop, and mobile environments while integrating more smoothly into different agent frameworks and deployment targets. Based on the Qwen family, Holo3.1 improves robustness across the environments where computer-use agents are actually deployed, addressing the distribution shifts that appear across mobile devices, alternative agent harnesses, and different execution frameworks. The release expands Holo3’s capabilities beyond browser and desktop control, with major gains in mobile automation, including AndroidWorld improvements from 67% to 79.3% for the 35B-A3B model and from 58% to 71% for the smaller 4B and 9B variants. Holo3.1 also introduces native support for function-calling protocols in addition to structured JSON outputs, helping teams deploy the model inside third-party agent stacks with near-parity between function-calling and native execution. -
5
Gemini 3.5 Live Translate
Google
Gemini 3.5 Live Translate is Google’s latest audio model for live speech-to-speech translation, delivering near real-time translation in more than 70 languages. The model automatically detects multilingual input and generates smooth, natural-sounding translated speech that preserves the speaker’s intonation, pacing, and pitch. Unlike turn-by-turn translation systems that wait for someone to finish speaking before responding, Gemini 3.5 Live Translate processes speech as it streams and generates translated audio continuously, balancing the need for context with the need to stay in sync. It stays only a few seconds behind the speaker throughout a session, helping conversations feel more fluid and natural, without awkward pauses. It is built for multilingual calls, meetings, lessons, broadcasts, live interpretation, dubbing, simultaneous translation, and voice translation applications. -
6
North Mini Code
Cohere
North Mini Code is Cohere’s first agentic coding model for developers and the inaugural member of its next generation of powerful models. Small, efficient, and open-source, it is built for the sovereign developer ecosystem and designed to deliver strong software development performance without requiring extensive hardware. North Mini Code is a mixture-of-experts model with 30B total parameters and 3B active parameters, giving developers access to agentic coding capabilities in a compact and efficient form. The model is optimized for code generation, agentic software engineering, and terminal tasks, with a 256K total context length and up to 64K maximum generation. It is built for real-world developer workflows, including understanding and orchestrating sub-agents, mapping system architecture, running code reviews, and supporting coding agents that need to reason through complex software tasks. -
7
Cartesia Sonic-3.5
Cartesia
Sonic 3.5 is Cartesia’s fastest, most natural text-to-speech model, built for expressive, real-time voice generation with sub-90ms latency and native support for 42 languages. It is designed to follow transcripts faithfully, voice confirmation codes, and heteronyms correctly without preprocessing, and stay expressive enough to carry a real conversation. It supports languages intended to deliver native-quality speech. Sonic 3.5 focuses on clean audio across every language and voice, with no artifacts to edit out, making it practical for production voice experiences where quality, speed, and consistency matter. Its expressive conversational delivery provides strong pacing and real emotional range, tuned for support and agent transcripts. Alphanumerics such as order numbers, phone numbers, IDs, and emails are spoken naturally in every language, while context-aware English pronunciation helps words like read, bass, and bow land correctly from the surrounding text. -
8
Cartesia Ink 2
Cartesia
Ink 2 is Cartesia’s fastest, most accurate streaming speech-to-text model, built for production voice agents with the lowest word error rate and best turn detection of any streaming STT. It is designed to transcribe structured data such as phone numbers, dates, and emails correctly the first time, while also knowing when a speaker starts and finishes without requiring a separate voice activity detection system. Turn detection is built directly into the model, so voice agents can react to events instead of managing raw transcript segments. Ink 2 emits a full lifecycle of turn events, giving an agent clear signals for when to listen, interrupt, think, prepare a reply, cancel a premature response, or speak. The transcript property is cumulative within a turn, meaning each update contains the full text transcribed so far rather than a delta, and emitted text is final once sent. -
9
SubQ 1.1 Small
Subquadratic
SubQ 1.1 Small is a long-context AI model from Subquadratic designed to reason over complete enterprise artifacts such as codebases, document collections, contracts, and financial filings. It uses Subquadratic Sparse Attention, or SSA, to reduce the high compute costs normally associated with processing very large context windows. The model delivers near-perfect long-context retrieval across 1M, 2M, 6M, and 12M token tests while using far less attention compute than dense attention. SubQ 1.1 Small also maintains strong general reasoning, coding, knowledge, and agentic task performance across multiple benchmarks. Its capabilities make it useful for financial analysis, legal review, contract work, software engineering, due diligence, and other workflows where information is spread across large artifacts. SubQ is built for organizations that want to move beyond fragmented retrieval pipelines and enable direct reasoning over massive bodies of information. -
10
Seedance 2.5
ByteDance
BytePlus Seedance provides official access to Seedance 2.5, a next-generation AI video generation model for creating professional AI video from text, image, audio, and video inputs. Seedance 2.5 adopts a unified multimodal audio-video joint generation architecture, giving creators comprehensive content reference and editing capabilities for highly controlled video creation. It supports text-to-video, image-to-video, and multimodal generation workflows, allowing users to transform ideas, images, reference clips, and audio cues into cinematic video outputs. Built for immersive audiovisual creation, Seedance 2.5 features strong motion stability and audio-video joint generation, helping produce ultra-realistic scenes with more natural movement and synchronized sound. The model is designed for director-level control, supporting images, audios, and videos as references so creators can guide performance, lighting, shadow, camera movement, scene direction, and visual style. -
11
HappyHorse 1.1
Alibaba
HappyHorse 1.1 is an upgraded AI video generation model designed to improve professional content creation across short dramas, ecommerce advertising, brand marketing, CG, and cinematic storytelling. The model enhances motion expressiveness, subject consistency, multi-reference fusion, instruction following, visual quality, and audio performance. HappyHorse 1.1 produces smoother actions, stronger kinetic tension, more natural pacing, and better temporal consistency in complex scenes. It also improves the preservation of product details, brand elements, character identity, storyboard references, and multi-panel inputs. The model delivers more realistic imagery, refined skin detail, stronger camera language, improved lip sync, richer sound design, and better audio-visual alignment. HappyHorse 1.1 helps creators, developers, and enterprise teams generate more controllable, coherent, and production-ready AI videos. -
12
Big Pickle
OpenCode Zen
Big Pickle is an AI model available through OpenCode Zen, a curated model provider focused on coding-agent workflows. The model is designed for text-based input, reasoning tasks, function calling, and developer workflows that require long-context understanding. Big Pickle supports a large context window, making it useful for working across bigger codebases, project files, technical prompts, and multi-step coding tasks. It can be accessed through OpenCode Zen using an OpenAI-compatible API format, allowing developers to integrate it into agentic coding tools and automation workflows. The model is positioned as a free or low-cost option within OpenCode’s coding-agent ecosystem. Big Pickle helps developers experiment with AI-assisted coding, reasoning, tool use, and long-context automation without relying only on premium frontier models.Starting Price: Free -
13
Ming-Flash Omni 2.0
Ant Group
Ming-Flash Omni 2.0 is a full-modal large language model from Ant Group, built on a unified multimodal architecture with “modal unity + task unity” as its core design philosophy. As part of the Ming series, it is designed to achieve cross-modal understanding and generation across text, images, audio, and video, allowing one model to see, hear, speak, and draw instead of relying on multiple specialized models. Ming-Flash Omni 2.0 follows the evolution of Ming-Light Omni and Ming-Flash Omni Preview, moving from unified architecture validation and hundred-billion-parameter scaling to a Data Scaling strategy that achieves open-source SOTA performance on multiple benchmarks. The model integrates four core capability modules: image-text understanding, video analysis, speech synthesis, and image generation or editing. For image-text understanding, Ming introduces structured knowledge graphs for fine-grained visual perception. -
14
Nano Banana 2 Lite
Google
Nano Banana 2 Lite is Google’s fastest Gemini Image model in the Nano Banana family, built for high throughput, speed, and scale. Also known as Gemini 3.1 Flash Lite Image, it is designed for rapid ideation and high-velocity developer pipelines where speed, iteration, and efficient production are the primary constraints. Developers can use it as the recommended replacement for the first version of Nano Banana, gaining immediate benefits across key performance dimensions while continuing to build image-generation and editing workflows through Google AI Studio, the Gemini API, and Gemini Enterprise Agent Platform. Nano Banana 2 Lite is optimized for near-real-time, high-volume workflows where ultra-low latency is critical, delivering text-to-image outputs in just a few seconds and making it well-suited for interactive prototyping, visual drafting, creative exploration, and large-scale image generation. -
15
LongCat-2.0
LongCat
LongCat-2.0 is a 1.6 trillion total-parameter Mixture-of-Experts language model built on AI ASIC superpods, with about 48 billion parameters activated per token and strong performance across coding and agentic tasks. It is a substantial step up from previous LongCat models, combining large-scale sparse architecture with dedicated post-training for real-world software engineering, tool use, long-context reasoning, and multi-step agent workflows. LongCat-2.0 is trained and deployed entirely on AI ASIC superpods, with pretraining spanning more than 35 trillion tokens and millions of accelerator-hours, demonstrating frontier-scale training on alternative hardware platforms. To strengthen long-horizon tasks, the model introduces LongCat Sparse Attention and is trained on hundreds of billions of tokens of 1M-context data, giving it native support for ultra-long context tasks and reliable long-document understanding. -
16
Seed Audio 1.0
BytePlus
Seed Audio 1.0 is a non-streaming audio generation API based on HTTP, designed to generate complete audio from text prompts, reference audio, or reference images. It supports text-only generation, where audio is created directly from the prompt; reference-audio generation, where uploaded reference clips guide the output; and reference-image generation, where an image reference can be passed to generate audio from the text to be synthesized. Built as part of BytePlus Seed Speech, Audio 1.0 uses the seed-audio-1.0 model version and is positioned as an audio creation capability rather than a standard speech-only endpoint. It can generate voice, music, and sound effects in a single pass, making it useful for producing richer audio scenes without separately creating and mixing every track. The API is intended for developers building audio generation into applications, workflows, and production systems, with a request-based structure that lets teams submit prompts. -
17
GPT-Live
OpenAI
GPT-Live is a new generation of voice models for natural human-AI interaction, now powering ChatGPT Voice. It is built to make talking with AI feel much more like having a real conversation through a full-duplex architecture, meaning it can listen and speak at the same time. During conversations, GPT-Live can show it is paying attention with short acknowledgments like “mhmm” or “yeah,” engage in quick back-and-forth, or stay quiet when the user needs a moment to think. Instead of processing separate turns one after another, GPT-Live continuously processes input while generating output, allowing it to decide many times per second whether to speak, keep listening, pause, interrupt, or invoke a tool. For questions that require web search, deeper reasoning, or more complex work, GPT-Live can delegate to a frontier model behind the scenes and bring the result back into the conversation when it is ready, while still maintaining the flow of the voice interaction. -
18
GPT-Live-1
OpenAI
GPT-Live-1 is one of the two new GPT-Live voice models rolling out to ChatGPT users globally, built to make talking with AI feel much more like having a real conversation. It is powered by a full-duplex architecture, so it can listen and speak at the same time instead of waiting for one rigid turn to end before the next begins. During conversations, GPT-Live-1 can show it is paying attention with short acknowledgments, engage in quick back-and-forth, pause when the user needs a moment to think, or stay quiet when asked to listen. It continuously processes input while generating output, allowing the model to decide many times per second whether to speak, keep listening, pause, interrupt, or invoke a tool. GPT-Live-1 also separates natural interaction from deeper work: when a question requires web search, reasoning, or more agentic capabilities, it can delegate the task to a frontier model behind the scenes and bring the result back when it is ready. -
19
GPT-Live-1 mini
OpenAI
GPT-Live-1 mini is one of the two GPT-Live voice models rolling out to ChatGPT users globally, designed to bring more natural, intelligent, and responsive voice interaction to everyday conversations. Built with the same full-duplex approach as GPT-Live, it can listen and speak at the same time instead of waiting for rigid turn-by-turn exchanges. The model continuously processes input while generating output, allowing it to decide many times per second whether to speak, keep listening, pause, interrupt, or invoke a tool. This makes conversations feel faster, smoother, and more natural, with active listening, quick back-and-forth, better timing, and fewer awkward interruptions when the user pauses to think. GPT-Live-1 mini also benefits from the new ChatGPT Voice experience, where users can interrupt with a question, ask ChatGPT to slow down, or tell it to stay quiet and listen. -
20
Muse Image
Meta
Muse Image is Meta’s image generation model from Meta Superintelligence Labs, built into Meta AI for creating, editing, and sharing high-quality visuals. The model can turn simple conversational prompts into detailed images, blend multiple photos together, remove unwanted objects, generate legible text inside visuals, and create styled outputs such as portraits, posters, stickers, room redesigns, infographics, and fantasy scenes. Muse Image uses advanced reasoning through Muse Spark to plan layouts, understand context, look up real-time web information, and combine visual references more intelligently. Users can start with suggested presets, mention Instagram accounts to personalize creations, and sketch or annotate edits directly on top of an image. The model powers creative experiences across Meta AI, Instagram Stories, WhatsApp chats, and soon Facebook, Messenger, and advertiser tools through Meta Advantage+ creative. -
21
Muse Video
Meta
Muse Video is Meta’s upcoming video generation model from Meta Superintelligence Labs, previewed alongside the launch of Muse Image. The model is built on the same pretraining foundation as Muse Image and is designed to generate high-fidelity videos with native audio support. Muse Video focuses on prompt adherence, visual realism, temporal consistency, and the ability to create short scenes with clear motion, continuity, and audio context. It can generate a wide range of video styles, including cinematic footage, UGC-style ads, animal scenes, product commercials, handheld point-of-view clips, and realistic moments with sound effects, voices, and music. Meta is continuing to improve areas such as audio-video synchronization and physically accurate fast motion before broader release. Coming soon to creators and Meta AI, Muse Video is positioned as a powerful tool for generating dynamic media across Meta’s creative ecosystem. -
22
Seedream 5.0 Pro
ByteDance
Seedream 5.0 Pro is a multimodal image creation model built for advanced reasoning, efficient content creation, and professional production. In real production environments, visual appeal is only the starting point; what matters is whether the model can efficiently meet complex creative demands, close the gap between the creator’s intent and the final visual output, and deliver true usability. Compared to previous versions, Seedream 5.0 Pro improves image-text alignment, structural coherence, text rendering, and visual aesthetics, while introducing core breakthroughs in complex information visualization, interactive precision editing, realistic imagery, portrait textures, and native multilingual generation. It can accurately transform data, concepts, and dense text into professional layouts for high-density content production, including infographics, educational images, technical drawings, UI designs, posters, and specialized professional visuals. -
23
Bonsai 27B
PrismML
Bonsai 27B is the new multimodal flagship of the Bonsai family and the first 27B-class model to run on a phone. Based on Qwen3.6 27B, it brings a new capability tier to local devices: multi-step reasoning, structured tool calls, vision tasks, and computer-use agentic loops that stay coherent across many steps. Bonsai 27B comes in two variants. Ternary Bonsai 27B uses ternary weights with FP16 group-wise scaling, giving 1.71 effective bits per weight and a 5.9 GB footprint for the quality-oriented laptop-class version. 1-bit Bonsai 27B uses binary weights with the same group-wise scaling, giving 1.125 effective bits per weight and a 3.9 GB footprint that fits within the memory budget of an iPhone 17 Pro. Both variants run end-to-end across the language network, embeddings, attention, MLPs, and LM head with no higher-precision escape hatches. They are multimodal, with a compact 4-bit vision tower, so on-device workflows can understand screenshots, documents, and camera input. -
24
Seed2.1 Pro
ByteDance
Seed2.1 Pro is a next-generation AI productivity model built to handle complex, real-world work across general agents, code engineering, and multimodal understanding. It reliably executes multi-step tasks for high-value office work and everyday consultation, including project planning, file processing, research, tool use, spreadsheet analysis, lesson-plan slide generation, and industry report creation across tools and environments. In software development workflows, Seed2.1 Pro strengthens end-to-end delivery by improving requirement understanding, architecture design, coding, debugging, implementation, and validation. Its agent capabilities are designed to make steady progress on difficult tasks and return practical, verifiable results rather than isolated responses. The model also advances knowledge, reasoning, visual understanding, spatial reasoning, and long-context processing, giving agents a stronger foundation for complex decision-making and execution. -
25
Seed2.1 Turbo
ByteDance
Seed2.1 Turbo is a next-generation AI productivity model designed to execute complex real-world tasks with strong general-agent, coding, and multimodal capabilities. It goes beyond one-off answers by carrying multi-step workflows toward defined goals and producing practical, usable outcomes across tools, environments, and interaction modes. For professional work and everyday consultation, it can support project planning, document and file processing, information analysis, solution design, content planning, tool use, and results consolidation. It also handles teaching, office, and research scenarios such as generating lesson-plan slides, analyzing complex spreadsheets, and producing industry reports. In software engineering, Seed2.1 Turbo supports end-to-end delivery across requirement analysis, feature implementation, bug fixing, environment setup, terminal usage, and result validation, while understanding codebase architecture, dependencies, and business logic to coordinate changes. -
26
Seeduplex
ByteDance
Seeduplex is a native full-duplex speech large language model built on a new “listen while speaking” framework for more natural, fluid, and precisely paced voice interaction. Unlike traditional half-duplex systems that alternate between listening and replying, it continuously receives and understands user-side audio, allowing it to listen and speak simultaneously while tracking the broader acoustic environment. Its high-precision interference suppression distinguishes genuine user interaction from background noise, broadcasts, navigation prompts, side conversations, and overlapping voices, reducing false responses and false interruptions in complex settings. Seeduplex also combines speech and semantic features for adaptive endpoint detection, helping it recognize when a user is thinking, hesitating, correcting themselves, or has actually finished speaking. It can wait patiently through reflective pauses, respond quickly once an utterance ends, and stop smoothly when interrupted. -
27
Gemini 3.5 Flash Cyber
Google
Gemini 3.5 Flash Cyber is a specialized cyber-focused model built on Gemini 3.5 Flash and fine-tuned to find, validate, and fix cybersecurity vulnerabilities efficiently at scale. It is designed for defensive security workflows where organizations need to identify critical weaknesses faster and generate reliable patches before those issues can be exploited. Flash’s combination of performance and efficiency makes it a strong foundation for scanning code, reasoning about security flaws, validating whether findings are real, and proposing targeted remediations across large software environments. Within CodeMender, multiple Gemini 3.5 Flash Cyber agents work together and combine their findings into a single report, helping the system investigate vulnerabilities from different angles and improve the quality of the final result. This coordinated agent setup delivers competitive frontier performance on CyberGym, a benchmark for evaluating cybersecurity capabilities. -
28
Laguna XS 2.1
Poolside
Laguna XS 2.1 is an upgraded open weight agentic coding model designed for long-horizon work on a local machine. It uses a 33-billion-parameter Mixture-of-Experts architecture with 3 billion activated parameters per token, retaining the same efficient architecture as Laguna XS.2 while improving multilingual software engineering and terminal-style task performance. The model is built to support coding agents that inspect repositories, reason through complex changes, use tools, execute commands, and continue working across extended tasks. It is served with a 256K context window, giving agents room to work with large codebases, lengthy histories, and multi-step workflows. Laguna XS 2.1 is supported by vLLM, SGLang, NVIDIA TensorRT-LLM, Hugging Face Transformers, and Ollama, with native llama.cpp support planned. It is available in BF16, FP8, INT4, and NVFP4 checkpoints, allowing developers to choose between maximum fidelity and configurations suited to tighter VRAM or compute budgets. -
29
Antares
Cisco
Antares is a family of open-weight security small language models purpose-built to localize known vulnerabilities inside large codebases. Antares-350M and Antares-1B are compact enough to run locally or on premises, helping teams keep proprietary source code inside their environment while reducing inference cost and runtime. Starting from a vulnerability description, advisory, or CWE category, the model follows an iterative investigation process similar to a human analyst, it searches for relevant code patterns, reads candidate files, incorporates new evidence, changes direction when a path is unproductive, and narrows the search to the files most likely to contain the weakness. Antares returns a ranked list of potentially vulnerable source files together with the terminal exploration trace that produced the result, making findings easier to review and prioritize. -
30
BLOOM
BigScience
BLOOM is an autoregressive Large Language Model (LLM), trained to continue text from a prompt on vast amounts of text data using industrial-scale computational resources. As such, it is able to output coherent text in 46 languages and 13 programming languages that is hardly distinguishable from text written by humans. BLOOM can also be instructed to perform text tasks it hasn't been explicitly trained for, by casting them as text generation tasks.
