Map of major players, model families, and the open source vs closed source debate.
The AI landscape changes too quickly for a static list of model numbers. This module gives you a current map of the leading labs, how their model families fit together, and a durable way to choose between them.
No single ranking captures the whole market. These selected labs are influential across frontier capability, open-weight releases, developer adoption, and product reach.
GPT-6 Astra · GPT-6.1 Sol · GPT-6 Luna
The GPT-6 family has three tiers: Astra is OpenAI's most capable model for the most demanding work, GPT-6.1 Sol offers near-Astra performance at a lower cost, and Luna is its most efficient model for focused, high-volume tasks.
Key strength: Agentic coding, long-context knowledge work, multimodality, and a broad product ecosystem.
Official model information ↗Claude Fable 5.1 · Opus 5.5 · Sonnet 5.5 · Haiku 5.5 · Mythos 5.1 (limited)
Claude has four broadly available tiers. Fable 5.1 is the top model for demanding reasoning and long-horizon agent work, and Opus 5.5 is Anthropic's suggested starting point for most workloads. Mythos 5.1 is the same model as Fable 5.1 with different safeguards, offered only through vetted access programmes.
Key strength: Long-running agents, coding and analysis, safety research, and the Model Context Protocol ecosystem.
Official model information ↗Gemini 4 Argon · 3.8 Flash · 3.6 Flash · 3.5 Flash-Lite · 3.1 Pro (preview)
Gemini 4 Argon is Google's flagship reasoning model, with a one-million-token context window. Flash and Flash-Lite models cover faster, lower-cost work, and current Gemini text models accept images, audio, video, and PDFs alongside text.
Key strength: Multimodal systems, scientific research, broad infrastructure, and on-device deployment.
Official model information ↗Grok 4.7 · Grok 4.6 · Grok 4.5
Grok 4.7 is the newest Grok model, described in its documentation as a frontier model for coding, agentic tasks, and knowledge work. Current Grok models accept text and image input.
Key strength: Large-scale compute, software engineering workflows, and integration with the X ecosystem.
Official model information ↗Kimi K3 · Kimi K2.6
Kimi K3 is Moonshot's 2.8-trillion-parameter flagship for long-horizon coding, knowledge work, and reasoning. It has native vision and a one-million-token context window, and its weights were released on 27 July 2026 under a custom Kimi K3 licence.
Key strength: Long-context agentic work, coding, native vision, and a fast-moving open-model programme.
Official model information ↗GLM-5.3 · GLM-5.2
GLM-5.3 is Z.ai's latest flagship, with a one-million-token context window and text-only input, released under a custom GLM-5.3 licence. The earlier GLM-5.2 remains available under the MIT licence.
Key strength: Open frontier models, agentic coding, long context, and multilingual capability.
Official model information ↗Mistral Large 4 · Large 3 · Medium 3.5 · Small 4
Mistral's catalogue mixes downloadable models under Apache 2.0 or modified MIT licences with managed commercial offerings. Mistral Large 4 is its newest model, with open weights promised, and its Devstral 2 and Magistral models were retired from the API on 31 July 2026.
Key strength: Efficient architectures, multilingual performance, specialised models, and European deployment options.
Official model information ↗Muse Spark Open · Muse Glimmer · Llama 4 Scout · Llama 4 Maverick
Llama 4 introduced natively multimodal mixture-of-experts models and remains a common base for self-hosted AI. Meta's newer open-weight models use the Muse name, including Muse Spark Open, released on 15 October 2026 under the custom Muse Model License.
Key strength: Open-weight distribution, research, community adoption, and a broad deployment ecosystem.
Official model information ↗Model names differ by vendor, but most families still serve four practical needs. Choose the least expensive tier that reliably completes your task, then evaluate with your own examples.
| Tier | Purpose | Current examples | Typical use |
|---|---|---|---|
| Frontier | Maximum capability for the hardest work | GPT-6 Astra, Claude Fable 5.1, Gemini 4 Argon, Kimi K3 | Long-running agents, research, complex engineering |
| High capability | Deep reasoning with production flexibility | GPT-6.1 Sol, Claude Opus 5.5, Grok 4.7, GLM-5.3 | Analysis, coding, planning, enterprise workflows |
| Balanced | Strong quality at useful speed and cost | Claude Sonnet 5.5, Gemini 3.8 Flash, Mistral Medium 3.5 | Everyday knowledge work and production applications |
| Fast / light | Low latency and high throughput | GPT-6 Luna, Claude Haiku 5.5, Gemini 3.5 Flash-Lite, Mistral Small 4 | Classification, extraction, routing, high-volume tasks |
"Open source" is often used loosely in AI. A more precise distinction is whether you can download the model weights or must access the model through a provider.
Llama 4 and Meta's Muse models, Kimi K3, GLM, Qwen, DeepSeek, Mistral models, Google's Gemma, and OpenAI's gpt-oss
GPT-6, Claude 5, Gemini 4, and Grok 4.7
Foundation models are one layer in a much larger stack. Real-world systems also depend on compute, orchestration, evaluation, data infrastructure, and user-facing applications.
OpenAI Developers
Official list of OpenAI's current flagship and specialised models.
Anthropic
Current Claude model family, context windows, and official capability guidance.
Google AI for Developers
Official index of stable, preview, and experimental Gemini models.
Moonshot AI
Official release overview for Kimi K3 and its architecture.
Rowan Cheung
A daily overview of AI product launches and industry moves.
Nathan Benaich & Air Street Capital
An annual report covering AI research, industry, policy, and safety.
7 questions · Score 70% or higher to complete this module
You can retake the quiz as many times as you need. Your best score is saved.
