OpenAI, Anthropic, Google release new AI model tiers with distinct pricing

OpenAI released the GPT-5.6 family on July 9, replacing a single model with three capability tiers: Sol (flagship for hard coding and research at $5 input / $30 output per 1M tokens), Terra (balanced for everyday production at $2.50 / $15), and Luna (speed and price for high-volume workloads at $1 / $6). The API adds programmatic tool calling, persisted reasoning, prompt-cache controls, and beta multi-agent orchestration. Anthropic followed with Claude Sonnet 5 on June 30, designed as a sensible default for coding and agents at scale. Claude Fable 5, restored July 1 after a temporary suspension, targets difficult software engineering and research at $10 per 1M input and $50 per 1M output tokens. Both Anthropic models offer a workhorse vs. flagship choice similar to OpenAI's approach.

Google introduced Gemini 3.5 Flash on May 19, prioritizing long-horizon agentic work, coding, multimodal understanding, and fast output. It's available in the Gemini app, AI Mode, API, and enterprise products. Gemini 3.5 Pro is still in development, so Flash is the current 3.5 family member. Across all labs, the advice is clear: start with the cheapest model that passes your own evaluation, then move up only when task failures justify the extra cost. For most coding and agent workflows, Claude Sonnet 5 or GPT-5.6 Terra offer the best balance. For frontier research, GPT-5.6 Sol and Claude Fable 5 lead. Speed-focused teams should evaluate Gemini 3.5 Flash.