Opus 5 and Fable 5 Lead DiG-bench Tier 7 With 0.2 Task Wins
Import AI 469 from August 17, 2026 places Opus 5 and Fable 5 with Claude Code ahead of GPT-5.5 in overall DiG-bench standings, and reports that only those two models logged any Tier 7 task wins, each at 0.2. That figure offers builders a tangible read on frontier coding agents facing the newsletter's hardest autonomous-research tier, alongside science AI coverage and an RSI simulator on the same trajectory. The issue also frames Meta strategy as casting superintelligence as decentralized individual empowerment rather than centralized utility, a shift toward foundational capabilities rivaling OpenAI and Anthropic under democratized superintelligence rhetoric. A 0.2 win ceiling suggests early Tier 7 capability, and the packet omits Tier 7 task makeup, statistical confidence, and complete Meta analysis.
Opus 5 and Fable 5 Lead DiG-bench Tier 7 With 0.2 Task Wins
Opus 5 and Fable 5 with Claude Code are the best overall models, followed by GPT-5.5. Only Opus 5 and Fable 5 were able to beat any tasks (0.2) in (Tier 7).
Key takeaway
Opus 5 and Fable 5 with Claude Code lead DiG-bench Tier 7, the only models to register task wins at 0.2—a narrow benchmark signal on the path to autonomous researchers.
What happened
In Import AI 469, published August 17, 2026, the headline highlights Opus 5 and Fable 5 leading DiG-bench Tier 7 with 0.2 task wins—the only models able to beat any tasks on that tier.
The newsletter body ranks Opus 5 and Fable 5 with Claude Code as the best overall models, followed by GPT-5.5, and frames the broader AI frontier around developing capable autonomous researchers, alongside science AI and an RSI simulator.
Evidence
Opus 5 and Fable 5 lead DiG-bench Tier 7 with 0.2 task wins.
Import AI · attributed
Opus 5 and Fable 5 Lead DiG-bench Tier 7 With 0.2 Task Wins
Only Opus 5 and Fable 5 were able to beat any tasks on Tier 7.
Import AI · attributed
Only Opus 5 and Fable 5 were able to beat any tasks (0.2) in (Tier 7).
Opus 5 and Fable 5 with Claude Code rank as the best overall models, ahead of GPT-5.5.
Import AI · attributed
Opus 5 and Fable 5 with Claude Code are the best overall models, followed by GPT-5.5.
The new frontier of AI is developing capable autonomous researchers.
Import AI · attributed
The new frontier of AI is developing capable autonomous researchers
Meta is positioning superintelligence as a decentralized engine for individual economic empowerment and creative agency.
Import AI · attributed
Meta is positioning superintelligence not as a centralized utility but as a decentralized engine for individual economic empowerment and creative agency.
Meta is pivoting from social graph optimization to foundational AI capabilities competing with OpenAI and Anthropic.
Import AI · attributed
This represents a significant strategic pivot for Meta, moving from social graph optimization to foundational AI capabilities that directly compete with OpenAI and Anthropic on the premise of 'democratized' superintelligence.
Why it matters
For builders evaluating coding agents for research workflows, Tier 7's 0.2 win ceiling suggests frontier models are only beginning to crack hardest autonomous research benchmarks, while Meta's democratized superintelligence framing signals intensifying competition in consumer agent infrastructure.
Limits and uncertainties
DiG-bench Tier 7 task wins at only 0.2 indicate limited demonstrated success on the hardest tier, and the packet does not define Tier 7 task composition or statistical confidence.
Meta strategic analysis in the packet is truncated and may not capture the full Import AI 469 argument on Zuck's technological pessimism.
Practical implications
Teams benchmarking coding agents should compare Opus 5, Fable 5 with Claude Code against GPT-5.5 specifically on DiG-bench Tier 7 before assuming parity on autonomous research tasks.
Builders targeting consumer agent ecosystems should track whether Meta's decentralized superintelligence positioning yields shipped creation tools and personal agent APIs.
What to watch
Whether Opus 5 or Fable 5 improve beyond 0.2 task wins on DiG-bench Tier 7 in subsequent benchmark releases.
How Meta's decentralized superintelligence positioning translates into shipped consumer agent APIs and creation tools.
