LLM provider
An organization that builds and/or serves large language models for use via API or open weights — the umbrella concept of this wiki. The 2026 field has four overlapping kinds:
- Proprietary frontier labs — OpenAI (GPT-5.x), anthropic (Claude Opus/Sonnet, claude-opus-4-8), google (Gemini), xAI (Grok). API-only; lead reasoning leaderboards (llm-benchmarks).
- Open-weight labs — deepseek, Meta (Llama), Alibaba (Qwen), Moonshot (Kimi), Z.ai (GLM), Mistral, Microsoft (Phi). Ship downloadable weights — see open-weight-models.
- Cloud resellers / platforms — Azure OpenAI, AWS Bedrock; host others’ models for regulated/enterprise buyers.
- Inference-as-a-service — run open weights behind OpenAI-compatible APIs (the niche deepseek also competes in on price).
The axes that differentiate them
- Access model — proprietary API vs open weights vs hosted.
- Cost — a ~600× spread; see llm-api-pricing.
- Capability/speed — multi-axis, see llm-benchmarks.
- Licensing — Apache-2.0 / MIT vs custom/community (open-weight-models).
- Compatibility — converging on OpenAI-compatible (and increasingly Anthropic-compatible) API shapes to lower switching costs.
Cross-wiki
Anthropic’s role as the substrate under the tools-for-thought ecosystem is tracked in research-wiki (anthropic, claude-opus-4-8); how these models run efficiently is llm-inference-wiki (llm-inference). This spoke owns the market/landscape.
Related
open-weight-models · llm-api-pricing · llm-benchmarks · deepseek · anthropic