← All articles

Claude Opus 4.8 Takes #1 on the Artificial Analysis Index — and Cuts Fast Mode 3×

AI News Published June 15, 2026 · anthropic · claude opus 4.8 · llm benchmarks · artificial analysis index · api pricing

On May 28, 2026, Anthropic released Claude Opus 4.8 across claude.ai, the Claude API, and Claude Code, callable as claude-opus-4-8. The headline result, confirmed by the independent benchmarking firm Artificial Analysis, is that Opus 4.8 became the first model to clear 60 on its aggregated Intelligence Index, scoring 61.4 and taking the #1 slot from OpenAI's GPT-5.5.

The interesting part is not that Anthropic shipped another frontier model — it is the shape of the improvement and the pricing move underneath it. This was, by Anthropic's own framing and that of several reviewers, a modest but tangible step rather than a generational jump, paired with an aggressive cut to the cost of serving the model in real time.

The Benchmark Picture

The Artificial Analysis Intelligence Index (v4.0) aggregates ten evaluations spanning reasoning, coding, science, and agentic tool use: GDPval-AA, τ²-Bench Telecom, Terminal-Bench Hard, SciCode, AA-LCR, AA-Omniscience, IFBench, Humanity's Last Exam, GPQA Diamond, and CritPt. Opus 4.8's 61.4 sits +4.1 points above Opus 4.7 (57.3) and +1.2 points ahead of GPT-5.5 at its xhigh effort setting (60.2), the previous index leader.

The wider gaps appear on agentic coding, which is where Anthropic has concentrated its competitive energy. Anthropic's own evaluation data puts Opus 4.8 at 69.2% on SWE-Bench Pro versus 58.6% for GPT-5.5 — a roughly 11-point lead on a benchmark that measures whether a model can resolve real software-engineering tasks. On GDPval-AA, a measure of economically valuable knowledge work, Anthropic reported a launch score of 1,890 at the model's "max" effort setting, +137 points over Opus 4.7 and +121 ahead of GPT-5.5 xhigh.

Read the index honestly: a +1.2 aggregate-point lead is a single-percentage-point edge on a composite metric, and many of the supporting figures — especially SWE-Bench Pro and GDPval-AA — are vendor-reported. The Artificial Analysis ranking is third-party and credible, but the deeper coding claims warrant independent replication before a team rebuilds a pipeline around them. A one-point composite lead does not, on its own, mean a different product-level experience.

The Pricing Move That Actually Matters

Standard pricing for Opus 4.8 held flat at the Opus 4.7 rate: $5.00/M input tokens and $25.00/M output tokens. Holding price while raising capability is itself notable in a market where frontier launches have often arrived with premium price tags.

The more consequential change is Fast Mode. It serves the same Opus 4.8 model at approximately 2.5× the standard speed, and Anthropic priced it at roughly one-third of the previous generation's fast tier — reported at $10/M input and $50/M output, down from the prior $30/$150. For interactive, latency-sensitive products — chat surfaces, IDE-resident agents, customer-facing flows — that cut changes the calculus of whether you can afford to put a frontier model in the hot path at all.

Conjecture, marked clearly: a 3× cut to fast-tier pricing alongside flat standard pricing reads less like a sudden efficiency windfall and more like a deliberate land-grab for real-time, high-frequency workloads — exactly the surface where latency, not raw benchmark score, decides which model a product ships. Anthropic appears to be pricing to win the interactive tier specifically, where switching costs are high and per-call volume is enormous. Teams should still benchmark Fast Mode's quality on their own tasks: "same model, faster" is the marketing claim, and serving optimizations can subtly shift output characteristics versus the standard tier.

Shipped the Same Day as a $965B Raise

Opus 4.8's launch was not an isolated product event. On the same day, May 28, 2026, Anthropic announced a $65 billion Series H at a $965 billion post-money valuation, led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital. That makes Anthropic the most valuable private AI company in the world, sitting just under the $1 trillion threshold no private company has crossed, and — per Bloomberg's reporting — eclipsing rival OpenAI's valuation.

Anthropic framed the raise as funding to "advance our research and expand our capacity to meet growing demand for Claude," and reporting cited a revenue run-rate climbing toward roughly $47 billion annualized, with October 2026 floated as a target IPO window. The Fast Mode price cut fits that narrative: capacity expansion is precisely what lets a lab serve a frontier model at 2.5× speed without rationing it behind a punitive price.

What It Means for Builders

For most teams, the practical takeaways are concrete. If you are choosing a frontier model for agentic coding, Opus 4.8's SWE-Bench Pro lead is the strongest single argument in its favor — but validate it on your own repositories, not the leaderboard. If you run a real-time, interactive product, Fast Mode at a third of the old price is the headline feature, not the +1.2 index points. And if you are simply tracking the frontier, the signal is that the gap between the top two labs has compressed to roughly a single composite point — the race is close enough that task-specific evaluation matters far more than which name sits atop the index this month.

Frequently asked

What did Claude Opus 4.8 actually score on the Artificial Analysis Intelligence Index?

Opus 4.8 scored 61.4 on the Artificial Analysis Intelligence Index (v4.0), making it the first model to clear 60 by a clean margin. That is +4.1 points above Claude Opus 4.7 (57.3) and +1.2 points ahead of GPT-5.5 at its xhigh setting (60.2), the previous index leader. The index aggregates ten evaluations including Humanity's Last Exam, GPQA Diamond, SciCode, Terminal-Bench Hard, and τ²-Bench Telecom.

How much does Claude Opus 4.8 cost?

Standard pricing held flat at the Opus 4.7 rate: $5 per million input tokens and $25 per million output tokens. The notable change is Fast Mode, which Anthropic priced at roughly one-third of the prior generation's fast tier — reported at $10/M input and $50/M output, down from $30/$150 — while running at approximately 2.5× the standard model's speed for latency-sensitive interactive workloads.

Is the +1.2 point lead over GPT-5.5 meaningful?

It is real but narrow. Anthropic and several outlets characterized 4.8 as a 'modest but tangible' improvement rather than a generational leap. The larger gap shows up on agentic coding: Anthropic's own data puts Opus 4.8 at 69.2% on SWE-Bench Pro versus 58.6% for GPT-5.5. As always, many of the headline figures are vendor-reported; independent replication is the definitive check before relying on a 1-point aggregate lead.

What is Fast Mode and when should you use it?

Fast Mode is the same Opus 4.8 model served at roughly 2.5× throughput for latency-sensitive, interactive use — chat surfaces, IDE autocomplete-style agents, and customer-facing flows where time-to-first-token matters more than squeezing maximum reasoning depth. It costs more per token than standard but a third of the previous fast tier, which materially changes the economics of running a frontier model in a real-time product.

How does this fit Anthropic's $965B valuation?

Opus 4.8 shipped on May 28, 2026, the same day Anthropic announced a $65 billion Series H at a $965 billion post-money valuation — the most valuable private AI company in the world, just under the never-before-crossed $1 trillion private threshold. The model and the raise together signal Anthropic is funding capacity expansion to meet enterprise demand, with reporting flagging October 2026 as a target IPO window.

Sources & further reading

Last reviewed June 15, 2026. AI Pulled News is editorial; corrections welcome at /news/about.html.