Claude Opus 5 Is #1 Overall. It Still Loses at the One Thing I Run Daily.

2026-07-27 · Source: LLM on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, quick

Summary

Claude Opus 5 has achieved the #1 overall ranking on BenchLM.ai with a score of 85.88, surpassing GPT-5.6 Sol's 81.46 across 215 models. Opus 5 demonstrates a "generational jump" in raw coding ability, scoring 96.0% on SWE-bench Verified, and ranks #3 of 129 in agentic tasks. Despite its high capabilities, Opus 5 lacks an equivalent to GPT-5.6 Sol's "ultra mode," which boosts Terminal-Bench 2.1 scores from 88.8% to 91.9% by fanning out tasks across parallel subagents. Furthermore, Opus 5's single \$5/\$25 per million token pricing tier is less flexible than GPT-5.6's three tiers (Sol at \$5/\$30, Terra at \$2.50/\$15, Luna at \$1/\$6), though GPT-5.6 has a billing catch for requests exceeding 272K input tokens.

Key takeaway

For AI engineers evaluating LLMs for daily workflows, especially those involving numerous simple tasks or complex agentic orchestration, you should look beyond overall benchmark scores. Prioritize models that offer specialized execution modes, like GPT-5.6 Sol's "ultra mode" for parallel processing, and flexible, tiered pricing structures. This approach ensures cost-effectiveness and optimal performance for your most frequent operational needs, rather than defaulting to the highest-ranked model.

Key insights

Overall benchmark leadership does not guarantee superiority for specific, common daily workflows.

Principles

• Raw model capability does not equate to practical utility.
• Specialized execution modes significantly enhance task performance.
• Tiered pricing models enable cost optimization for varied workloads.

Method

GPT-5.6 Sol's "ultra mode" splits tasks across parallel subagents, improving Terminal-Bench 2.1 scores from 88.8% to 91.9%.

In practice

• Utilize GPT-5.6 Sol's "ultra mode" for parallelizing large jobs.
• Leverage GPT-5.6's Luna tier for cost-effective simple tasks.
• Monitor GPT-5.6 input tokens to avoid higher billing rates above 272K.

Topics

• LLM Benchmarking
• Claude Opus 5
• GPT-5.6 Sol
• Agentic AI
• LLM Pricing
• SWE-bench
• Terminal-Bench

Best for: AI Architect, NLP Engineer, CTO, AI Engineer, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• OPUS 5 CLICK NOW — Claude Opus 5 redefines LLM value, offering superior performance at a significantly lower cost per task.
• [D] What is even the point of these LLM benchmarking papers? — LLM benchmarking papers often become obsolete quickly due to rapid model updates, but their datasets can still be valuable.
• I Tested GPT-5.6 Sol vs Claude Sonnet 5 on Real Agents. The Brutal Truth. — Optimal AI model selection for production agents hinges on matching model strengths to specific workload demands, not just overall capability.
• The Two Best AI Models/Enemies Just Got Released Simultaneously — New LLMs like Claude Opus 4.6 offer significant productivity gains but introduce complex ethical and reliability challenges.
• The Only LLM Comparison Guide You Need in 2026 — In 2026, LLM selection shifts from "best overall" to "best for specific tasks, cost, and deployment."
• Opus 5 — Claude Opus 5's practical coding and agentic capabilities appear to exceed its initial aggregate benchmark scores, highlighting evaluation challenges.

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 LLM on Medium.