Version 2

GPT-5.6 launched July 9 as a three-tier family (Sol, Terra, Luna) with OpenAI claiming Sol leads Claude Fable 5 by 13.1 points on Agents' Last Exam at roughly one-quarter the cost [1][2]. Anthropic's Fable 5 (launched June 9) is priced at $10/$50 per million tokens versus Sol's $5/$30; Anthropic also released Claude Mythos 5, the same model with cybersecurity safeguards removed for vetted researchers [3][2]. Benchmark comparisons are contested: OpenAI published an audit on launch day finding ~30% of SWE-Bench Pro tasks broken and retracting its prior recommendation — the same benchmark where Fable 5 outscored Sol 80% to 64.6% [5][2]. Independent tester Simon Willison found Sol 'competent' but 'not clearly superior' to Fable for complex coding tasks [2].

The two models now compete on both capability and cost, with Sol substantially cheaper at comparable or better performance on OpenAI's preferred benchmarks. The benchmark integrity dispute — each lab's preferred metrics favor its own model — means there is no neutral scorecard for choosing between them. Meta's Muse Spark 1.1 at $0.80/M input tokens [6] adds further cost pressure and may erode the two-model frontier framing if it proves competitive.

GPT-5.6 launched officially on July 9, 2026 as a three-tier family: Sol ($5/$30 per million tokens), Terra ($2.50/$15), and Luna ($1/$6) [1][2]. Claude Fable 5 is priced at $10/$50 per million tokens [3]. OpenAI's launch benchmarks place Sol 13.1 points ahead of Fable 5 on Agents' Last Exam (53.6 vs. 40.5) and 2.8 points ahead on the Artificial Analysis Coding Agent Index at less than half the output tokens [1]. Sol also scores 62.6% on OSWorld 2.0 and 73.5% on ExploitBench2, the latter a large jump from GPT-5.5's 47.9% [1]. A new 'ultra' mode runs four parallel agents by default for demanding tasks [1]. GPT-5.6 was simultaneously announced as the new default model in Microsoft 365 Copilot across Word, Excel, PowerPoint, and related tools [4].

On the same day as the launch, OpenAI published an audit concluding that roughly 30% of SWE-Bench Pro tasks are broken across four failure categories, and retracted its prior recommendation of the benchmark [5]. The timing drew immediate notice: Fable 5 outscored Sol on SWE-Bench Pro 80% to 64.6%, and Simon Willison observed directly that the audit publication 'may help explain why OpenAI chose to publish this article specifically calling out SWE-Bench Pro for problems they found while auditing that benchmark' [2]. OpenAI frames the audit as responsible transparency, noting frontier model pass rates on SWE-Bench Pro's public split rose from 23.3% to 80.3% in eight months — evidence the benchmark is saturated [5]. Willison adds from personal testing that Sol is 'definitely very competent, though so far it hasn't struck me as better than Fable at the kind of complex coding tasks' he uses [2].

Anthropic's Claude Fable 5 launched June 9 alongside Claude Mythos 5 — the same underlying model with cybersecurity and biomedical safeguards removed for vetted research partners in Project Glasswing [3]. Fable 5 uses safety classifiers that fall back to Claude Opus 4.8 for sensitive queries in under 5% of sessions; Anthropic acknowledged this will produce false positives [3]. In extended autonomous testing, Mythos 5 conducted novel genomics research over more than a week, training a model that outperformed a recent paper in Science while being 100 times smaller [3]. The Fable 5/Mythos 5 tiered architecture — Mythos access restricted to vetted partners under mandatory 30-day data retention — is a different safety posture than OpenAI's general availability of Sol with layered but uniform safeguards [1].

The competitive landscape now includes a third significant entrant: Meta's Muse Spark 1.1 at $0.80 per million input tokens, applying direct pricing pressure on both labs [6]. Grant Harvey at The Neuron characterized the situation as 'Anthropic's game to lose,' given rivals offering frontier-class models at lower prices [6]. OpenAI's rebranding of Codex as 'ChatGPT for Work' — a super-app combining web browsing, file editing, scheduling, and multi-agent coordination — generated widespread confusion; both Ethan Mollick and Willison said they could not distinguish it from standard ChatGPT [6]. One documented cost-optimization strategy: using Fable 5 as a planning advisor with cheaper Sonnet executing tasks achieves roughly 92% of Fable's benchmark performance at about 63% of the cost [6].