OpenAI moved GPT-5.6 Sol, Terra, and Luna to general availability on July 9, roughly 24 hours after SpaceXAI opened Grok 4.5 to the public. The Artificial Analysis Coding Agent Index now has a leader that isn't Anthropic, and the frontier vendors have quietly agreed that coding-agent benchmarks are the metric that matters.

Sol at max reasoning scores 80 on the Coding Agent Index, 2.8 points above Anthropic's Fable 5, "while using less than half the output tokens, taking less than half the time, and costing about one-third less," per OpenAI's own writeup. Turn reasoning down to medium and Sol still beats Fable 5 by 11.4 points at roughly one-quarter the estimated cost. Terra reportedly edges Fable 5 as well; Luna outperforms Opus 4.8. The pricing runs $5/$30 per million input/output tokens for Sol and $1/$6 for Luna.

Grok 4.5, which SpaceXAI trained in partnership with Cursor, ships at $2 input and $6 output per million tokens. Opus 4.7 sits at $5 and $25. On Datacurve's DeepSWE 1.0, Fable at max leads with 66.1%, GPT-5.5 xhigh follows at 64.31%, Grok 4.5 lands at 62.0%, and Opus 4.8 max trails at 55.75%. The more interesting data point is SWE-Bench Pro token efficiency: Grok 4.5 resolves the average task in 15,954 output tokens against Opus 4.8's 67,020. A 4.2x gap.

The harness matters too. On DeepSWE 1.1's mini-swe-agent, the ranking inverts: Fable max at 70%, GPT-5.5 xhigh at 67%, Grok 4.5 at 53%. Coding agents don't have a single score anymore; they've a matrix, and vendors are learning to cite the cells that flatter them.

The regulatory frame is worth naming. For a two-week window before Wednesday, OpenAI restricted GPT-5.6 to what CNBC characterized as a "small group of trusted partners" whose participation was shared with the U.S. government, running through the Commerce Department's Center for AI Standards and Innovation with METR as external evaluator. OpenAI ran over 700,000 A100-equivalent GPU hours of automated red teaming under its Preparedness Framework. The company also made clear it "doesn't believe this kind of government access process should become the long-term default," a line that reads less like a policy position and more like a marker being laid down for the next administration cycle under President Trump.

The pattern rhymes with the 2018 cloud-computing FedRAMP moment, when hyperscalers accepted a bespoke federal review regime as a cost of doing business, then spent the following years quietly renegotiating its scope. Frontier labs are entering the same phase. Grok 4.5 hits the EU in mid-July. The Coding Agent Index has a new number one, and it took roughly a day.

Sources

https://openai.com/index/gpt-5-6/
https://x.ai/news/grok-4-5
https://techcrunch.com/2026/07/09/openai-launches-its-new-family-of-models-with-gpt-5-6/
https://techcrunch.com/2026/07/08/spacexai-releases-grok-4-5-which-elon-describes-as-an-opus-class-model/
https://www.cnbc.com/2026/07/08/openai-expanding-gpt-5point6-ai-model-release-ending-government-limits.html