Anthropic

The AI safety company behind Claude. Anthropic fields a clean three-tier closed lineup — Opus (most capable), Sonnet (balanced default), Haiku (fast and cheap) — known for coding quality and a safety-first posture.

Anthropic is an AI-native company founded in 2021 by a group of former OpenAI researchers, structured as a public-benefit corporation with an explicit focus on AI safety. Its single product line is Claude, a family of models offered in three tiers across successive generations. Amazon and Google are major investors, and Claude is available not only through Anthropic's own API but also via Amazon Bedrock and Google Cloud Vertex AI.

For business readers, the model is simpler than OpenAI's sprawling catalog: instead of dozens of variants, Anthropic offers three tiers per generation. Opus is the flagship for the hardest work, Sonnet is the balanced default most teams should use, and Haiku is the fast, cheap tier for high-volume tasks. Pick the tier that matches the job.

Anthropic is closed by design — there are no open weightsThe numerical values inside a trained model that encode everything it has learned. A model is, functionally, a giant list of weights — tens of billions of numbers for a mid-sized model, hundreds of billions for a frontier model. "Open-weight" means those numbers are published. "Downloading the weights" means getting the actual file you'd need to run the model yourself., and there's no equivalent of OpenAI's gpt-oss or Google's Gemma. You rent Claude through an API; you don't download or self-host it. The company frames this partly as a safety choice (retaining the ability to monitor and adjust deployed models), and partly it's the business model.

The lineup has a notable dynamic: capability creeps upward across tiers over time, so each new Sonnet generation tends to match or beat the previous generation's Opus on coding and document tasks — at Sonnet pricing. The practical implication is that the flagship is often overkill; the mid-tier frequently does the job. Anthropic's models are particularly associated with coding reliability, careful instruction-following, and long-context work (Opus and Sonnet carry 1M-tokenThe basic unit of text a model reads and writes. Tokens are roughly three-quarters of a word in English — so 100 tokens is about 75 words. Models don't see letters or words directly; they see tokens. Pricing is almost always quoted per million tokens, and context windows are measured in tokens rather than words. windows).

Two things matter most. First, it's closed and premium-priced: Claude pricing generally runs higher than OpenAI's comparable tiers, and there's no self-hosting option at all, so if owning the model or keeping everything fully in-house is a hard requirement, Claude isn't the fit — an open-weightA model where the trained weights are freely downloadable — you can run it yourself without contacting the creator. Llama, Mistral, Qwen, and Gemma are open-weight. Open-weight does not mean open-source: the training data and code often stay private. The license still governs what you can do with the weights, including whether you can use them commercially. line like Meta's Llama, Qwen, or Google's Gemma is. Second, the value lever is mostly Sonnet plus prompt caching: for the majority of workloads, Sonnet 4.6 with cached context is the practical default, with Haiku for routing and Opus reserved for work that justifies the premium.

On data governance, Anthropic is US-based with enterprise and zero-retention options, plus availability through Bedrock and Vertex for buyers who want it inside an existing cloud relationship. There are no open weightsThe numerical values inside a trained model that encode everything it has learned. A model is, functionally, a giant list of weights — tens of billions of numbers for a mid-sized model, hundreds of billions for a frontier model. "Open-weight" means those numbers are published. "Downloading the weights" means getting the actual file you'd need to run the model yourself., so the only deployment path is hosted.

Against the other closed Western frontier labs — OpenAI and Google — Anthropic competes most directly on coding quality and safety positioning, while OpenAI offers far more lineup breadth (and an open line in gpt-oss) and Google leads on native multimodality and ecosystem reach. Against the open-weightA model where the trained weights are freely downloadable — you can run it yourself without contacting the creator. Llama, Mistral, Qwen, and Gemma are open-weight. Open-weight does not mean open-source: the training data and code often stay private. The license still governs what you can do with the weights, including whether you can use them commercially. players (Meta, Qwen, DeepSeek, Gemma), the trade is stark: Claude offers no self-hosting at all, so those are the only options if open weightsThe numerical values inside a trained model that encode everything it has learned. A model is, functionally, a giant list of weights — tens of billions of numbers for a mid-sized model, hundreds of billions for a frontier model. "Open-weight" means those numbers are published. "Downloading the weights" means getting the actual file you'd need to run the model yourself. are required. The honest summary: Claude is a strong, premium, closed choice that many teams pay up for on coding and reliability — and a non-starter for anyone who needs to own the model.