VC
Value Add VC
⚡HomePulse⚡Helpful Apps📝Blog
← Value Add PulseAI4 of top 5 Intelligence Index slots

Anthropic Just Won the Benchmark War It Started

Weeks after a government-ordered shutdown over jailbreak concerns, Claude Fable 5 has returned to top independent AI benchmarks, with four Anthropic models occupying the top five Intelligence Index slots.

#1, score 60
Fable 5 Rank
>99%
Jailbreak Block Rate
19 days
Suspension Length
TC
Trace Cohen
Early-stage VC & angel · Founder, New York Venture Partners
July 5, 2026
1 min read
ShareXLinkedInEmail
THE RUNDOWN
1

Claude Fable 5 sits at rank 1 on the Intelligence Index with a score of 60, its highest public benchmark standing since its June 12 export-control suspension

2

Anthropic's Opus 4.8 (56) and other Claude models occupy three more of the top five spots, with a single GPT-5.5 entry (55) the only non-Anthropic model in that tier

3

The suspension stemmed from an Amazon-discovered jailbreak technique; Anthropic's new safety classifier now blocks that specific technique in over 99% of attempts

4

Claude Sonnet 5, a more agentic model with stronger tool use and autonomous task handling, launched alongside Fable 5's return, broadening Anthropic's product lineup

TC
The VC Read · Trace's TakeTrace Cohen

A 19-day government-ordered shutdown over a real jailbreak vulnerability should have been a black eye; instead Anthropic used it to ship a better safety classifier and came back topping the leaderboard. That's a rare instance of a forced pause becoming a competitive advantage -- founders dealing with their own compliance-driven downtime should take note that the fix itself can become the differentiator.

Claude Fable 5's return to full global availability on July 1 has been followed by a clean sweep of independent AI benchmark rankings: the model now sits at rank 1 on the Intelligence Index with a score of 60, ahead of Anthropic's own Opus 4.8 at 56 and OpenAI's GPT-5.5 at 55, with Anthropic models occupying four of the top five spots overall.

The comeback is notable given how the suspension happened. The US government ordered Fable 5 and the more restricted Mythos 5 offline on June 12 after Amazon researchers found a jailbreak technique that could prompt the model into identifying software vulnerabilities it was supposed to refuse to discuss -- a genuine national-security concern, not a routine safety hiccup. Anthropic spent the 19-day suspension building an improved safety classifier specifically targeting that technique, which the company says now blocks it in over 99% of attempts, and that fix is what convinced the Commerce Department to lift the export controls on June 30.

Alongside Fable 5's restoration, Anthropic launched Claude Sonnet 5, described as its most agentic Sonnet model yet, with stronger reasoning, tool use, coding and autonomous task handling, rolled out across Claude Code, the Claude Platform and Claude Enterprise with introductory pricing and higher rate limits -- broadening Anthropic's product lineup at exactly the moment its flagship model is topping public leaderboards again.

The sequence -- shutdown, fix, benchmark-topping relaunch, new product tier -- is a genuinely unusual turnaround story: most companies forced into a 19-day government-ordered shutdown over a security vulnerability would expect to lose ground to competitors during the outage, not emerge with their flagship model ranked first.

What to watch: whether OpenAI or Google respond with their own releases aimed at reclaiming the top Intelligence Index spot in the coming weeks, and whether Anthropic's new safety classifier holds up against novel jailbreak attempts now that Fable 5 is back at global scale.

ShareXLinkedInEmail
More onAnthropic →

Originally reported by Anthropic. Analysis and editorial commentary by Value Add Pulse.

← Back to Pulse

THE WIRE in your inbox

Tech, startup & VC news with Trace's take. Free, no spam.

Read Next

AIFable 5: #1 on Intelligence Index

Anthropic, Not OpenAI, Is Setting AI's 2026 Agenda

Between topping independent benchmarks, courting Samsung on custom silicon and lining up an October IPO near $1.1 trillion, Anthropic is dictating the pace of the AI race while OpenAI plays defense in Washington.

AI2 frontier customers reportedly in talks

Samsung's 2nm AI Chip Talks Now Include Meta Too

Samsung's custom 2nm AI chip discussions, first reported with Anthropic, have reportedly widened to include Meta, positioning the Korean foundry as a serious third option against TSMC for frontier AI silicon.

AIOpen-weight model: July 2026 access

Mistral Is Building Its Own Anti-OpenAI Playbook

Mistral AI is preparing an open-weight model with July early access and its own Nvidia-powered compute platform, positioning the French lab as Europe's clearest sovereign alternative to OpenAI and Anthropic.

@Trace_Cohen·t@nyvp.com