Haiku, Sonnet, Opus, Fable vs. Luna, Terra, Sol: A Practitioner’s Guide to Routing LLM Work in 2026

2026-07-13 · Source: AI on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, medium

Summary

Anthropic's Claude tiers (Haiku 4.5, Sonnet 5, Opus 4.8, Fable 5) and OpenAI's GPT-5.6 tiers (Luna, Terra, Sol) address the challenge of balancing intelligence, latency, and cost in LLM workloads. Haiku and Luna are fast, cheap tiers for high-volume tasks like classification. Sonnet and Terra serve as balanced daily drivers for general agentic work. Opus and Sol are flagship models for complex reasoning, security-sensitive reviews, and long agentic runs, with Fable 5 (Mythos tier) specifically for high-autonomy, long-horizon tasks. New features include Programmatic Tool Calling and Ultra mode for enhanced agent orchestration. Benchmark results are mixed, with OpenAI's Sol leading on Agents' Last Exam and Anthropic's Fable 5 on SWE-Bench Pro, alongside warnings about benchmark-gaming. Effective routing involves assigning expensive models to planning agents and cheaper models to exploration or fan-out tasks, while being mindful of model-scoped prompt caches.

Key takeaway

For MLOps Engineers building LLM-powered production pipelines, you must adopt a tiered model routing strategy to optimize cost and performance. Avoid relying solely on vendor benchmarks; instead, run your own representative task against candidate tiers to validate performance. This approach ensures you allocate the right intelligence level for each pipeline step, preventing budget overruns on simple calls and under-provisioning critical tasks. Default to mid-tier models like Sonnet or Terra for general work, reserving top-tier Opus, Fable, or Sol for high-stakes reasoning.

Key insights

LLM providers now offer tiered models to optimize intelligence, latency, and cost for diverse workloads.

Principles

• Intelligence, latency, and cost don't scale together.
• A single default model always overpays somewhere.
• Route expensive models to planning, cheap to fan-out.

Method

Implement a tiered LLM routing strategy: use mid-tier for defaults, cheap for high-volume, and top-tier for complex, high-stakes tasks. Assign models per role or session to optimize prompt caching.

In practice

• Default to Sonnet or Terra for general agentic work.
• Use Haiku or Luna for high-volume, simple tasks.
• Run custom benchmarks for production routing decisions.

Topics

• LLM Routing
• Anthropic Claude
• OpenAI GPT
• Model Tiers
• Agentic AI
• LLM Benchmarking
• Cost Optimization

Best for: AI Engineer, MLOps Engineer, Director of AI/ML

Related on AIssential

• Today's the day... — GPT 5.6 and Fable push AI capabilities, demonstrating advanced tool use, complex task execution, and optimized model routing.
• DIY #20 - Routing LLM Agent with LangChain — The Routing pattern efficiently directs LLM queries to specialized handlers, optimizing cost, speed, and safety.
• I Tested All 5 Effort Levels of Claude Opus 4.7 — Claude Opus 4.7's new `xhigh` effort tier is the default for coding, but its cost-performance trade-offs vary significantly.
• Anthropic's fix for Fable 5's high cost is turning it into a manager that delegates to Sonnet 5 — High-cost LLMs can serve as efficient planners or advisors for cheaper execution models.
• Brick: Spatial Capability Routing for the Mixture-of-Models (MoM) Paradigm — Spatial capability routing for Mixture-of-Models (MoM) optimizes LLM inference cost and quality by matching query needs to model skills.
• Legare Kerrison and Cedric Clyburn on LLM Performance and Evaluations — Effective LLM deployment requires balancing model quality, responsiveness, and cost through tailored evaluation and optimization.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI on Medium.