Anthropic vs Cohere

Anthropic's cheapest option at $1.25/M tokens (likely Claude Haiku) includes vision and reasoning capabilities across all 13 models, while Cohere's $0.150/M Command models lack these features entirely. The 28-point performance gap between top models (Claude Sonnet 4.6 at 66/100 vs Command R+ at 38/100) suggests Anthropic optimizes for capability density over price competition.

Cohere focuses on text-only workflows with just 2 of 4 models supporting function calling and zero vision support, while Anthropic provides 100% vision coverage and 85% reasoning coverage across 13 models. This specialization allows Cohere to offer prices 66x lower on the high end ($10 vs $150/M tokens) but limits them to traditional NLP tasks.

The 1M token context enables processing entire codebases or document collections in single prompts, critical for enterprise RAG systems where Anthropic's 92% web search capability (12/13 models) outperforms Cohere's 0%. At enterprise scale, the $148.75/M token premium for Anthropic's top tier often costs less than the engineering required to chunk workflows for Cohere's 256K limit.

Cohere's Command R+ at $10/M tokens offers 15x cost savings over Anthropic's comparable mid-tier options while maintaining 50% function calling support, making it viable for high-volume classification or extraction tasks. The single open source model in Cohere's portfolio also enables on-premise deployment, impossible with Anthropic's fully proprietary stack of 13 models.