Anthropic Claude Fable 5 tops frontier benchmarks; Opus 4.8 now #2 at same Opus 4.6 pricing

<cite index="42-1,42-4">Claude Opus 4.8 (May 28, 2026) topped the Artificial Analysis Intelligence Index at 61.4, edging out GPT-5.5, with standout gains in coding (SWE-bench Pro 69.2%) and computer-use agents (Online-Mind2Web 84%).</cite> <cite index="45-2,45-3">Claude Fable 5 entered in July 2026 as Anthropic's first generally available Mythos-class model, leading WebDev Arena at 1653 Elo with a 92-point gap over #2—the widest recorded gap ever.</cite> <cite index="47-2">Fable 5 holds 95% SWE-bench Verified, 11 points ahead of Opus 4.8 on the Pro variant (80.3% vs 69.2%), at 2x the output token cost ($10/$50 vs $5/$25).</cite>

<cite index="44-1,44-2">Opus 4.8 scored 1,890 on GDPval-AA with 'max' effort, +121 points ahead of GPT-5.5 xhigh, implying a ~67% win rate against OpenAI's second-ranked model while using 15% fewer turns per task and 35% fewer output tokens than Opus 4.7.</cite> <cite index="44-4">Claude now leads on Humanity's Last Exam scientific reasoning and ranks #2 on AA-Omniscience at 27.4 (behind only Gemini 3.1 Pro at 32.9), with Anthropic maintaining substantially lower hallucination rates at 35.9% versus peer models.</cite>

<cite index="45-2">Opus 4.8 replaces Opus 4.7 at unchanged pricing of $5/$25 per million tokens, with Opus 4.7 dropping to #5 after a single-month tenure as top-ranked model—the largest monthly shake-up since tracking began.</cite> For enterprises evaluating architecture, the Fable 5 vs Opus 4.8 trade-off is now explicit: pay 2x per token for 11-point SWE-bench Pro gain and wider domain coverage (Fable 5), or use Opus 4.8 as the cost-optimized production default with equivalent GDPval leadership. The frontier ranking will likely remain volatile through Q3 2026 as GPT-5.6 Terra and Sol broaden distribution and xAI's Grok 4.5 enters general availability.