Companies, models, apps. Let’s untangle them.
A model is the engine; an app is the place you use it. The name on a chat window doesn’t tell you everything about what’s running underneath.
This is a dated map, not a buying recommendation or a complete ranking. “Open weights” means parameters are downloadable under stated terms; it does not automatically mean unrestricted or fully open source.
| Company / family | Models & focus | Access | What to check |
|---|---|---|---|
| OpenAI | GPT-6 Astra · GPT-6.1 Sol · GPT-6 Luna A range spanning demanding reasoning, balanced work, and smaller focused tasks. | Hosted API / supported ChatGPT and Codex experiences | Availability varies by product, plan, client, and workspace. API usage and app subscriptions have separate charging rules. Source / provider ↗ |
| Anthropic | Claude Haiku 5.5 · Sonnet 5.5 · Opus 5.5 The newest 5.5 releases span small, middle, and larger model classes. Fable and restricted Mythos lines also exist. | Hosted Claude / API | A family name is not a benchmark. Compare a specific version and effort setting. Source / provider ↗ |
| Gemini 3.8 Flash · Gemini 4 Argon (restricted) Google’s public API catalog lists Gemini 3.8 Flash and separate audio models. Argon’s September announcement describes a restricted rollout. | Hosted / restricted Gemini / developer platforms | Do not assume a frontier announcement is already available in your personal app. Source / provider ↗ | |
| Grok / xAI | Grok 4.7 The current release emphasizes coding and knowledge work. | Hosted Grok / API | Product browsing, connectors, and the underlying model are separate parts of the experience. Source / provider ↗ |
| Meta | Muse Wonder · Muse Spark · Muse Glimmer · Llama Meta’s model families span hosted and downloadable offerings. An October 12 external evaluation covers Wonder and Wonder Full. | Mixed Meta products / model downloads | Wonder coverage here is an external evaluation briefing; confirm current provider access and the exact download license. External evaluation ↗ |
| Alibaba / Qwen | Qwen3.8 · Qwen3.8-27B The 27B variant is the local collaborator used in this draft. The family also contains much larger models. | Open weights Qwen / local runtimes / API | Our samples use MLX 5-bit quantization and a 16k loaded context; other setups will vary. Source / provider ↗ |
| DeepSeek | V4.1 Flash The release supports image and text understanding; the API changelog documents routing changes. | Open weights DeepSeek / API / compatible runtimes | Open weights can still require large-scale infrastructure. Record the actual model version. Source / provider ↗ |
| Mistral | Large 4 preview · smaller Mistral families Large 4 is in public preview. Mistral’s changelog still says downloadable weights are coming; its Vibe product includes a Chat mode for the legacy Le Chat experience. | Preview / mixed Vibe / Mistral Studio | The Large 4 preview is not yet a verified downloadable local model in this draft. Source / provider ↗ |
| Moonshot | Kimi K3 A very large multimodal model with its own model license. | Open weights Kimi / API / compatible runtimes | Open-weight access is not a promise of personal-computer compatibility. Source / provider ↗ |
| Z.ai | GLM-5.3 The model card describes post-training improvements on the GLM-5.2 base. | Open weights Z.ai / API / compatible runtimes | Read the current model card and its license before deploying. Source / provider ↗ |
| Google / Gemma | Gemma 4 · EmbeddingGemma 2 A separate family from Gemini, with language and specialized embedding releases. | Open models Local runtimes / developer tools | An embedding model retrieves or represents information; it is not a replacement chat assistant. Source / provider ↗ |
| IBM | Granite 4.2 IBM’s August release includes 3B, 8B, and 30B language-model sizes. | Open models Compatible runtimes / enterprise tools | Verify runtime support and memory needs for the particular download. Source / provider ↗ |
| Cohere | Command A+ · North Small Translate Command focuses on enterprise tasks; the September blog also highlights a translation release. | Open / enterprise Cohere / model deployments | Check each model’s terms and intended purpose; commercial terms can differ by model. Source / provider ↗ |
| Reflection | Beam A newly announced coding and reasoning model, with final evaluations underway. | Preview · weights pending Early access | We have not verified public downloadable weights; do not plan a local installation yet. Source / provider ↗ |
| NVIDIA | Nemotron · 3.5 Lightning NVIDIA’s August 11 release pairs the open Nemotron 3.5 Lightning model with NeMo Switchyard for routing requests among models. | Open ecosystem NVIDIA / compatible runtimes | Provider speed claims and hosted benchmarks do not predict performance on your own hardware. Source / provider ↗ |
| Microsoft | Phi family Microsoft’s Phi page describes a family of smaller open models for a range of deployment settings. | Open models Azure / compatible runtimes | This is a model-family guide, not a verified latest-release claim. Check the particular model card and supported runtime. Source / provider ↗ |
| Ai2 | Olmo · Molmo · Bolmo Ai2 publishes open research artifacts across language, vision and specialized models. Its October 7 Bolmo update adds new byte-level checkpoints. | Open models Research downloads / compatible runtimes | Check the exact model’s purpose and release materials; openness has several dimensions. Source / provider ↗ |
| MiniMax | M3 · H3 · Speech 2.8 · Music 3.0 The catalog spans language, video, speech, and music. M3 has an official weight repository and a model-specific community license. | Mixed / open weights MiniMax apps / APIs / model downloads | Check the exact model’s terms, hardware needs, and custom loading code; not every product in the catalog is an open-weight release. Source / provider ↗ |
| Xiaomi | MiMo-V2.6-Pro-RL The official model card identifies this as the flagship checkpoint in the MiMo-V2.6 series. | Open weights MiMo / API / compatible runtimes | A large open checkpoint needs substantial infrastructure; this is not a laptop recommendation. Source / provider ↗ |
App subscriptions and API usage have separate charging rules. Access depends on the model, plan, region, and product. Links above are ordinary source links, not affiliate links.