Four new AI models in eight days, the AI race just reached a new level

Four AI launches almost back to back, here's what happened

The AI market just went through one of the most packed launch periods in its history. Four models came out within eight days, Grok 4.5 from SpaceXAI, GPT-5.6 from OpenAI, Muse Spark 1.1 from Meta, and Kimi K3 from Moonshot AI. For comparison, back in early June only two companies, Anthropic and OpenAI, had a model at this level at all. Now there are six.

We're talking about an independent ranking called the Artificial Analysis Intelligence Index. Think of it as a combined exam for AI models, one that pulls together several different tests and turns the result into a single scale. The higher the number, the smarter the model performs overall, not just at one narrow task.

Grok 4.5 launched on July 8 scoring 54 points. Just two days later, GPT-5.6 arrived in three versions at once, Sol scoring 59, Terra at 55, and Luna at 51, alongside Meta's Muse Spark 1.1 at 51. Then on July 16, Moonshot AI's Kimi K3 launched and immediately scored 57, landing third overall, ahead of even Claude Opus 4.8's 56 points.

How the leaderboard went from a duopoly to six players

^{Six labs now field a model above 50 on the Artificial Analysis Intelligence Index, Anthropic (Claude Fable 5, 60), OpenAI (GPT-5.6 Sol, max, 59), Moonshot AI (Kimi K3, 57), SpaceXAI (Grok 4.5, high, 54), Z AI (GLM-5.2, max, 51), and Meta (Muse Spark 1.1, xhigh, 51). Arrows mark the six entries from the past eight days.}

Before June 2026, only Anthropic and OpenAI had models scoring 51 or higher on this index. In mid-June, Z AI joined them with its GLM-5.2 model, scoring exactly 51. Then just last week, the list gained two more names at once, SpaceXAI with Grok 4.5 and Meta with Muse Spark 1.1. Kimi K3's launch made Moonshot AI the sixth company in this club, entering straight at 57 points.

It's worth noting that the top three models on the ranking now belong to three different companies, with just three points separating them. Four of the ten highest-scoring models launched only in the past eight days, and six of the ten launched since early June.

The one thing that hasn't changed is who holds first place. Claude Fable 5 has held the lead since June 9 with 60 points, but its margin over the closest competitor has shrunk from four points to just one. In effect, the race for the top spot has gotten so tight that almost any new release could shake it up.

What this race looks like on a longer timeline

^{The long view shows how unusual the past six weeks have been. For most of the period since late 2022, the frontier of the Intelligence Index was held by one or two labs at a time. Since early June, SpaceXAI, Moonshot AI, Meta, and Z AI have all closed to within single digits of number one.}

Looking at the ranking's full history since late 2022 makes it clear just how unusual the current situation is. Typically, one or two companies held the top of the table for extended stretches. Having four different labs close in on the leader all at once is something that hasn't happened before in this ranking's history.

Why Kimi K3 debuted straight into third place

Moonshot AI's Kimi K3 debuted with a strong result right out of the gate, and that's no accident. On GDPval-AA v2, a test that measures a model's performance on real professional tasks, Kimi K3 scores 1668 points on an Elo scale, similar to a chess rating, where the more confident wins against strong opponents, the higher the final score. That's the third-best result, behind only Claude Fable 5 at 1760 and GPT-5.6 Sol at 1748.

On a separate test, AA-Briefcase, which measures a model's ability to handle long professional tasks, Kimi K3 lands in second place with 1547 points, behind only Claude Fable 5's 1583, but ahead of GPT-5.6 Sol's 1495. Worth noting separately is the analytical quality score, where Kimi K3 hits 1760, effectively tied with Claude Fable 5's 1764.

Now for the price. Kimi K3 costs 3 dollars per million input tokens and 15 dollars per million output tokens, which works out to just 0.94 dollars per Intelligence Index task. For comparison, a comparably strong result from Claude Opus 4.8 costs 1.80 dollars per task, roughly twice as much for a similar level of intelligence.

How the coding leaderboard reshuffled

^{The launches also reshaped the Artificial Analysis Coding Agent Index. GPT-5.6 Sol in Codex now leads at 80, ahead of GPT-5.6 Terra in Codex at 77 and Claude Fable 5 in Claude Code at 77. Grok 4.5 in Grok Build scores 76, on par with GPT-5.5 in Codex, and Muse Spark 1.1 in Opencode enters at 69.}

These same four launches also shook up the separate coding-focused ranking. The new leader is GPT-5.6 Sol paired with the Codex tool, scoring 80 points. Just behind are GPT-5.6 Terra, also in Codex, at 77, and Claude Fable 5 paired with Claude Code, also at 77. Grok 4.5 paired with Grok Build scores 76, roughly on par with the previous GPT-5.5 in Codex. Muse Spark 1.1 paired with Opencode debuts at 69.

Why powerful AI got 2-3 times cheaper in just eight days

^{Eight days redrew the intelligence versus cost frontier. GPT-5.6 Luna ($0.21), Muse Spark 1.1 ($0.26), and Grok 4.5 ($0.31) all sit at or below the price GLM-5.2 set a week earlier at $0.32, while Kimi K3 ($0.94) and GPT-5.6 Sol ($1.04) deliver within three and one points of Claude Fable 5 ($2.75) at roughly a third of its cost per task.}

Perhaps the most interesting shift happened in pricing. GPT-5.6 Sol scores just one point below Claude Fable 5, yet costs 1.04 dollars per task versus the leader's 2.75 dollars. Grok 4.5 delivers a score of 54 for just 0.31 dollars, under a third of the price of the previous GPT-5.5 model, which cost 0.99 dollars for a comparable task.

At the 51-point level, the picture looks similar. GPT-5.6 Luna costs 0.21 dollars, and Muse Spark 1.1 costs 0.26 dollars, both cheaper than GLM-5.2's 0.32 dollars, which was the most affordable option at that intelligence level just a week earlier. In effect, AI nearly as powerful as the top models got two to three times cheaper in just eight days.

In the End

In just eight days, the balance of power in AI shifted more than it had over the previous several months combined. Six different companies now field high-level models, three different labs hold the top three spots, and prices for roughly the same level of intelligence dropped by two to three times. The leader, Claude Fable 5, still holds first place, but its margin has narrowed to a single point, and the next release could shake things up all over again.

For an everyday user, this means one simple thing, there's no longer a single "right" AI model for every situation. Grok 4.5, GPT-5.6, Kimi K3, Muse Spark 1.1, each model has its own strength and its own price, and figuring out which one fits your task best is something you basically have to redo every week.

Testing all six models separately, setting up access with each company on your own, takes too much time and isn't practical. It's a lot simpler when every current AI model is already gathered in one place, so you can compare results on your own task instead of just comparing leaderboard scores.

That's exactly what unitool.ai is for. Get one subscription and unlock access to all the newest models at once, including this week's fresh releases, no foreign cards and no separate sign-up on each company's site. See for yourself which model actually performs best on your specific tasks.