Opus 5.5 edges GPT-6 Astra on Artificial Analysis's rebuilt Intelligence Index, now 60% private tests

Today's lead is Artificial Analysis's rebuilt Intelligence Index. It leads because the scores vendors quote next week can't be compared with the ones they quoted last week. Also today: an Anthropic engineering manager says he stopped giving coding interviews on his own teams, and a newsletter headline turned that into company policy. Anthropic's 40% government discount now reaches state and city governments. If you quote an Index score this week, check which version it came from.

What changed this week

Opus 5.5 edges GPT-6 Astra on Artificial Analysis's rebuilt Intelligence Index, now 60% private tests

Artificial Analysis, an independent benchmarking site, rebuilt its Intelligence Index, the ranking launch posts cite, and says 60% of it now runs on private or held-out test sets, "for the first time," to curb benchmark gaming. It added FrontierCode 2.0, the harder coding benchmark from Cognition, the company behind the Devin coding agent, and a science benchmark. Claude Opus 5.5 leads the overall Index at 58, one point ahead of GPT-6 Astra's 57. On FrontierCode 2.0, Opus 5.5 leads at 46%; GPT-6 Astra, GPT-6.1 Terra and Claude Sonnet 5.5 tie at 42%. Cognition's own page scores Opus 5.5 at 45.5% on the new version, down from 54.6% on the old one, and every listed model fell. The test got harder, not the models. An HN reader noted that Anthropic's models ran through Claude Code and OpenAI's through Codex, each vendor graded through its own agent tool.

Announcing Artificial Analysis Intelligence Index v5

Index v5 features 2 new benchmarks and 3 upgrades. For the first time, the majority of the Index is built on private datasets

Artificial AnalysisArtificial Analysis

An Anthropic engineering manager says he dropped coding interviews on his own teams

Felix Rieseberg, an Anthropic engineering manager, says he "stopped asking candidates" to do coding interviews on his own teams. The post makes no claim about Anthropic's hiring beyond them. His reason, verbatim: "There's no more point in giving you a little React challenge than there is in quizzing people about assembly," because "Claude is now better at both." He now scans portfolios for prototypes that show "taste, empathy, and principled thinking." Latent Space's AINews digest headlined it "Anthropic ends Coding Interviews," a claim the post doesn't make.

Hunger beats tenure: Advice for junior engineers

The coding interview is dead, at least for my team at Anthropic. I've seen enough to know that it no longer provides signal. I have a spicily optimistic take: I think there's a huge opportunity for junior engineers.

Felix RiesebergFelix Rieseberg

Anthropic's 40% government discount expands from federal agencies to states and cities

Anthropic is extending its 40% discount on Claude Enterprise, Claude for Government and the API past federal agencies to state, local, tribal and territorial governments too, with no minimum spend required. Anthropic says the discounts are "estimated to save the government $1 billion," no period stated, and that about 500 government bodies already use Claude. The one coding-specific example it names: Utah gave Claude Code to 34 developers across 10 state agencies, and one IT director used it to document 12 legacy repositories.

Expanding our partnership with GSA to bring Claude to all levels of government

Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.

AnthropicAnthropic

Also worth your time

• Early progress on accelerating drug development with Claude
• GitHub Copilot supports server-to-server authentication

Know someone who'd want this in their inbox? Forward it — that's how this grows. And if we got something wrong, or you think we buried the real story today, hit reply. A person reads every one.

The New Way is written with AI. It gathers the day's stories, checks them against their sources and drafts every summary. A person decides what runs and reviews every issue before we hit send.