OpenAI executives personally published step-by-step tutorials teaching developers how to swap Claude Code for GPT-5.6 in five minutes, turning the model's July 9 launch into an open declaration of war on Anthropic's developer turf. The tutorials went viral, drawing over 1.3 million viewers as OpenAI's top experts walked developers through installing a CLI proxy, authenticating for both Claude and Codex, linking to Claude Code, and creating an alias called "claudex" that routes requests to GPT-5.6 instead.
"Share the recipe. Everyone wants to know how to use GPT-5.6 Sol in CC," one developer posted. "We don't care about the harness." The defection campaign is backed by hard numbers.
On the Artificial Analysis Coding Agent Index, GPT-5.6 Sol scored 80 points, 2.8 points above Claude Fable 5, while using less than half the output tokens, completing tasks in less than half the time, and costing about one-third less. On Agents' Last Exam, a benchmark spanning 55 professional fields, Sol hit 53.6 points, beating Fable 5 by 13.1 points.
OpenAI launched three tiers at once: flagship Sol at $5 per million input tokens and $30 per million output tokens, balanced Terra at $2.50/$15, and cost-efficient Luna at $1/$6. Pricing matches GPT-5.5 levels.
Terra slightly outperforms Fable 5 on the coding index while Luna surpasses Opus 4.8, with both models running at roughly one-third the time and one-quarter the cost of Anthropic's equivalents. The model release came with a platform shakeup.
OpenAI merged the standalone Codex app into a revamped ChatGPT desktop app, creating what the company calls a "superapp" with built-in browser control, computer operation, and a new Work agent that mirrors Anthropic's Claude Cowork. The Codex brand is gone, but its engine now powers ChatGPT Work.
Sam Altman acknowledged the pricing pressure directly, saying "every enterprise now is thinking about spend." The GPT-5.6 family introduces an "ultra" mode that coordinates up to 16 parallel agents for demanding tasks, and a "max" mode that extends reasoning time beyond the standard xhigh setting. On cybersecurity benchmarks, GPT-5.6 Sol scored 73.5% on ExploitBench versus GPT-5.5's 47.9%.
On ExploitGym, it nearly doubled GPT-5.5's peak pass rate from 15.1% to 24.9%. To stress-test safety, OpenAI doubled the top reward in its Bio Bug Bounty Program to $50,000 for researchers who can jailbreak GPT-5.6 into generating banned biology content. What started as a short-term challenge for GPT-5.5 is now a permanent program.
Anthropic responded by extending free quotas for Claude Fable 5 by seven days, an unforced concession that OpenAI's camp framed as panic. One developer posted a threat aimed at Anthropic: "If this path is blocked, I'll reset your limit for you."
GPT-5.6 Sol with max reasoning comes within one point of Fable 5 on the Artificial Analysis Intelligence Index while completing tasks 61% faster at roughly half the estimated cost. On OSWorld 2.0, Sol scored 62.6%, surpassing Opus 4.8 while using 85% fewer output tokens.
OpenAI noted that Claude Fable 5 was excluded from its GeneBench Pro biology evaluation because it "does not answer advanced biology questions and refuses the majority of questions in this eval."
Early customers are already reporting workflow gains. Lovable's co-founder said the new models reduce build steps by roughly 25%, cut tool calls by 35% to 48%, and reduce stuck tasks by 15%.
Cognition's CEO Scott Wu called GPT-5.6 "a top-tier model that combines strong coding-agent performance with very strong cost efficiency."
OpenAI's user base is approaching 7 million, and the company is now soliciting outstanding GPT-5.6 demos with promises of mystery gifts for the coolest entries. The message to developers is clear: bring your workflows, leave the harness behind.
