Google might’ve made most of the technical breakthroughs that have led to the current AI revolution, but it’s watching a bevy of labs move past it in model intelligence.
Google’s slow slide on the closely watched Artificial Analysis Intelligence Index has reached a new low. The latest snapshot of the index — version 4.3.2, which aggregates results from ten evaluations including AA-Briefcase, GDPval-AA, Terminal-Bench 4.0, Humanity’s Last Exam and AA-Omniscience — places Google tenth among AI labs, with its best-ranked model, Gemini 3.8 Flash, sitting 14th on the model-level chart at a score of 41.
That’s a long way from the top. Anthropic’s Claude Fable 5.1 (max, with tools) and OpenAI’s GPT-5.6 Astral (max) are tied for first at 53, with Claude Opus a close third at 51.
Between Google and the leaders sit 13 models from at least eight rival labs — including Meta’s Muse Spark 1.3 at 48, Grok 4.7 at 46, Alibaba’s Qwen3.8 Max and Z.AI’s GLM-5.3 at 45, and a three-way tie at 44 between Grok 4.6, Step 3 Preview and Moonshot’s Kimi K3.
From Leaderboard Top To Mid-Table #
The drop continues a pattern that has been building for months. In July, Google fell out of the top 5 on the index for the first time, an ignominious milestone for the company that invented the Transformer architecture underpinning modern AI. By August, it had slipped to eighth place — a period that also saw reports that Gemini 3.5 Pro had been quietly canceled, leaving Google without a true frontier flagship even as competitors shipped theirs.
The intervening weeks have made things worse, not better. Meta, SpaceXAI, Alibaba, Z.AI, Step and Moonshot have all released or refreshed models that now rank above anything Google has on the market. The most striking detail on the current chart may be who Google is trailing: not just Anthropic and OpenAI, but a roster of Chinese labs — Z.AI, Alibaba, Moonshot, Step and DeepSeek — whose models now crowd the middle of the leaderboard at price points that undercut Western frontier models dramatically. Moonshot’s Kimi K3 has been trading blows with closed frontier systems on real-world coding workloads for months now.
An Index That Keeps Moving The Goalposts #
Part of the story is methodological churn. Artificial Analysis has updated the index aggressively this year — retiring its standalone Agentic Index in September and re-anchoring scores — which is why the raw numbers look different from launch-day coverage. When Gemini 3.8 Flash debuted in early September, it scored 59 on the then-current version of the index, a three-point jump over Gemini 3.7 Flash that was framed as Google closing in on frontier performance at a fraction of the price. On the current v4.3.2 snapshot, that same model reads 41, behind GLM-5.3 Flash and GPT-5.6 Terra, and ahead of only DeepSeek V4.1 Flash, GPT-5.6 Luna and DeepSeek V4 Pro in the visible field of 27 models drawn from 655 total.
The Price-Performance Shield #
To be fair to Google, the Intelligence Index tells only half the story. Gemini 3.8 Flash remains one of the cheapest credible models on the market at $0.75 per million input tokens and $3.75 per million output, and it has consistently landed close to the intelligence-versus-cost Pareto frontier — the set of models offering the best capability per dollar at their respective price tiers. For businesses running high-volume agentic workloads, that math could matter more than raw leaderboard position.
But the strategic problem is real. Google’s entire 2026 model strategy has effectively been executed through the Flash line — iterative, algorithmic updates shipped at a breakneck cadence — while the Pro-tier successor to Gemini 3.1 Pro remains absent from the leaderboard entirely. The company’s next genuine shot at climbing back up the ranking likely rests on Gemini 4.0, which reports suggest is still some way off.
What It Means #
It’s not as though all is lost for Google in AI. The company still owns Search, Android and Chrome, and Gemini remains deeply embedded across all three. But the Intelligence Index matters because it measures what enterprises and developers increasingly buy on: agentic coding, professional knowledge work and long-horizon reasoning. Every position Google cedes to Anthropic, OpenAI or a Chinese open-weights lab is a position those rivals solidify in exactly the workflows where AI spend is growing fastest.
Tenth place isn’t fatal. It is, however, a remarkable place to find the lab that started the modern AI era — and each new index update has, so far, only pushed it further from the top.