Chapter 1
On the independent Artificial Analysis Intelligence Index
On the independent Artificial Analysis Intelligence Index, rebuilt as version 5 at the end of October 2026 and the third scale since September, Google's Gemini 4 Carbon leads at 59, one point ahead of Anthropic's Claude Opus 5.5, with OpenAI's GPT-6 Astra at 57 and GPT-6.1 Sol at 55. Carbon, like Argon before it, went first to vetted security partners rather than to the public. The best model anyone can download scores 32, far further back than on the old scale. The price of a fixed level of capability has been collapsing for years, 9 to 900 times a year by Epoch AI's measurement, and it is still moving: Opus 5.5 took the lead at $4 and $20 per million tokens, a fifth below the model it replaced; OpenAI halved its mid-tier and small-model prices in September and again in October, when GPT-6.1 Terra arrived at $1 and $5; and Anthropic's Claude Haiku 5.5 sells at ten and fifty cents. Access is contested on both sides: June 2026 brought the first model-level US export controls, OpenAI, Meta and Google have shipped their strongest models to limited partners first, and in late September OpenAI was reported to have shelved a planned flagship, GPT-6.1 Astra, after it fell short in the company's own alignment tests.
Chapter 2
Six years of record show three arcs
Six years of record show three arcs. Capability: from GPT-3 behind a private-beta API to today's crowded frontier band, where six labs shipped models within a hair of each other in a single week. Price: the steepest cost curve in the industry's history — Epoch AI measures inference prices for fixed performance falling 9 to 900 times a year, and Stanford's AI Index puts GPT-3.5-equivalent capability at a 280th of its late-2022 price by late 2024. Access: the pendulum has swung from API-only gatekeeping through the open-weights insurgency — Llama first, then DeepSeek, Qwen and GLM — to government-imposed restriction, and, through the second half of 2026, part-way back toward opening. July brought the largest open-weight release on record, Moonshot's Kimi K3; August brought Alibaba publishing the weights of its 2.4-trillion-parameter flagship, the largest model anyone can download, and Zhipu following with GLM-5.3. Meta — which made open-weight frontier-chasing a corporate strategy with Llama 2 in 2023, then retooled its Muse line as proprietary — reversed part-way, releasing a small on-device model and pledging to open its flagship's weights. That pledge is unmet: into September Meta kept reaching the frontier with closed models, shipping the proprietary Muse Spark 1.3 while the promised open weights never appeared. Open-weight leadership meanwhile stayed in China: in September Xiaomi published the highest-scoring downloadable model, and the first large American challenger, Reflection AI's Beam, was announced in October and published its weights on October 16, scoring 30 on the index.
Chapter 3
The open-versus-closed gap is the story's live question
The open-versus-closed gap is the story's live question, and on the September scale it was twelve points between the closed leader and the best downloadable model; on the October rebuild, which leans on private and harder tests, it is about 27. Open weights are also not the same as open reach. Kimi K3's licence is close to MIT — commercial conditions that bite only above US$20M in revenue or 100M monthly users — but its training data and process are undisclosed, and Rahul Shome of the Australian National University told Nature the model is too large for personal devices and will probably need institutional investment to run. GLM-5.3, at 753 billion parameters with 40 billion active, ships under Zhipu's own licence rather than a standard open one. Alibaba's Qwen3.8-Max carries a similar catch, and a second one: what Alibaba actually published is a text-only, always-reasoning sibling of the model it sells. The UK AI Security Institute separately measures the best open-weight model at four to seven months behind the closed frontier on cyber capability. What is moving down-market is cheap hosted capability and, increasingly, local capability: Alibaba's Qwen3.8-27B, small enough to run four-bit on a 32-gigabyte laptop, scores 34, and Zhipu's MIT-licensed GLM-5.3-Flash — 320 billion parameters but only 18 active — scores 42 while running natively multimodal. A whole class of cyber-specialist models, and Tencent's 770-billion-parameter Hy4, still ships with no independent score at all.
Chapter 4
There is a second race behind the first
There is a second race behind the first, and it is about what can be checked rather than what can be built. Every number above comes from one published independent index, because vendor launch materials have repeatedly been shown to grade their own homework — and that index is itself a moving object. Rebuilt in September with harder tasks and more private test sets held back to stop labs training against them, it re-scored every model at once, and the reshuffling was not uniform: GPT-6 Astra, which had scored level with the model it replaced and behind two rivals, came out of the rebuild ahead of both. No model changed. The test did. Six weeks later it was rebuilt again, this time with most of its tests private, and the open models fell furthest. That is the honest condition of the field: where an independent index exists, it is the thing arguments are settled with, and it is revised faster than the models are; where none exists — a security model sold inside a vendor's own harness, a Chinese open-weight release benchmarked only by its maker — a capability claim is just a press release. The labs increasingly ship both kinds at once, and the outside scorecard, not the launch chart, is what separates the tested from the merely asserted.
And now
The scale that settles arguments about which model is best was rebuilt on October 27. Artificial Analysis moved 60% of its Intelligence Index onto tests the labs cannot see, added real-world coding and science tasks, and re-scored everything; the new scores are not comparable with the old. Claude Opus 5.5 led the new version at 58, ahead of GPT-6 Astra at 57.
A day later Google's Gemini 4 Carbon took the top at 59. It leads four of the index's nine tests but not coding or knowledge work, and it costs about a quarter less to run the index than Opus 5.5. Few can use it: Google is releasing it first to vetted cyber defenders, with no date for developers or the public, at a standard price of $4 and $20 per million tokens.
On October 26 OpenAI released GPT-6.1 Terra, its mid-tier model, at $1 and $5 per million tokens, against $2 and $12 for the version it replaces. The index scored it six points higher at a third of the cost per task.
Among models anyone can download, the rebuilt index widened the distance to the closed leaders, while Mistral and MiniMax published weights they had promised.
Where it stands today →