GLM-5.3 can exploit the period between a security fix becoming publicly available and the fix reaching end users (the patch gap) by analyzing the security fixes it can reverse engineer the underlying vulnerabilities and develop working exploits before users receive the patches

It's been a bit over 3 months since Kimi K3 was announced, and I made the prediction that, contrary to what Twitter hype merchants were saying, it wouldn't reach Claude Fable level, or even GPT-5.6 Sol. It didn't take a rocket scientist to arrive at the same conclusion — Moonshot said so themselves in their release blog post, kudos to them: “…its overall performance still trails … Claude Fable 5 and GPT 5.6 Sol…”. This turned out to be true — and now I have data from MORE THAN 90 benchmarks to corroborate the claim. Yes, you read that correctly — ninety benchmarks. All of these benchmarks were released after Kimi K3, so the possibility of overfitting is very low. If we compare K3 with GPT-5.5 and Opus 4.8 — released in late April and May, respectively — I can call them equal without hesitation. Kimi loses on as many benchmarks as it wins, and there's a solid group of benchmarks within a practical tie margin. The median difference in scores across the 40+ shared benchmarks is within 1% against both models, which is very close. It is really fair to say “Kimi K3 was the Opus 4.8 of open-source”. Now to the grim part. The fresh benchmarks don’t support parity with Sol or Fable: — Against Sol, Kimi lost on 62 out of 89 benchmarks. The median relative deficit was 14%, and on 30 benchmarks the gap was more than 20%. — Against Fable, Kimi managed to win just 3 benchmarks, with 2 more practical ties, and lost 35 of the 40. On 23 benchmarks, the relative deficit exceeded 20%. — If we focus on Agentic Coding benchmarks, Kimi loses 10/10 to Fable, trailing by a median of close to 30% (!) in relative terms. The “Fable-class” claims have aged badly. In my summer post, I wrote the following about Kimi: “if we gather another 30 benchmarks, we might see, say, 8 wins for Kimi, 5 ties, and 17 losses.” This turned out to be optimistic. Even if we compare it to GPT-5.6 Sol, it is more like 5 wins, 4 ties, and 21 losses per 30 benchmarks (proportions taken from all 89 benchmarks comparing these two). All of that, of course, is not to mention that since then (and before K3.1) we got the beautiful Astra, Fable 5.1, and even 3 Claude models from the 5.5 family, including the beast that is Opus. So, “on average,” proprietary models are still out on top, and the gap is significant. I'm not ready to say whether it has shrunk recently, but I'm planning to expand my analysis in the future so that we can see whether the gap widens or narrows. (As usual, disclaimer: please don't take this as “Kimi / DeepSeek is trash, and nobody should use them.” Check out the link below the post for the full details on my position.)

We’ve announced the Artificial Analysis Progress Index: a single capability scale for every language model since 2022, launched alongside Intelligence Index v5 The Intelligence Index is still our primary measure of current model intelligence. It uses the evaluations most relevant to developers today, so its mix changes as evaluations are added and retired, and scores from different versions can’t be compared. The Progress Index is built for comparisons across time. It uses item response theory, a statistical method used to score standardized tests, to estimate one general capability score for each model from the underlying tasks in all our evaluations, including those that are retired or excluded from the Intelligence Index. Use the Intelligence Index to choose a model today. Use the Progress Index to see how capability has changed over time. The Progress Index builds on prior work placing models on a single capability scale, in particular @EpochAIResearch's Epoch Capabilities Index and the capability scale @AnthropicAI published in its Claude Mythos Preview system card. Key takeaways: ➤ Since reasoning models arrived, the frontier has risen ~65 to 70 points a year, about double the 2023 pace: the highest score among released models rose 35 points in 2023 (GPT-3.5 Turbo 125 to GPT-4 Turbo 160), 67 in 2024 (to o1 at 228), 68 in 2025 (to GPT-5.2 at 296) and 56 so far in 2026 (to Gemini 4 Carbon at 352). ➤ Open weights models generally trail proprietary models by about 4 months: since January 2025, each new highest open weights score came 2-8 months (median ~4) after a proprietary model first reached it. DeepSeek V4.1 Flash 1017 is the highest-scoring model with released weights at 324, a level Claude Fable 5 reached in June 2026. Kimi K3.1 Pro (weights not yet released) scores 341, about two months after the closed frontier ➤ New frontier scores are being set more often: the highest score among released models was raised on 2 release dates in 2023, 4 in 2024, 9 in 2025 and 11 so far in 2026. That is about once every four weeks this year. ➤ OpenAI has held the top score most of the time since November 2022, but the frontier has become a close contest: OpenAI held the number one position for the first 27 months measured, but in 2026 especially there have been rapid changes in the leader between OpenAI, Anthropic, and now Google

GPT 6.1 Terra (xhigh) scores 28.1% on WeirdML, comparable to Fable 5.1. Terra is better at token efficiency, while Fable has a higher final score. Terra is much cheaper than any comparable models.

I've always believed that intelligence should come from simple, general building blocks that learn their own abstractions directly from raw data; in other words, the bitter lesson. I think we're finally seeing a real step towards AI that can learn from all data. Language modeling on raw bytes is a microcosm of this ideal, and to me it's a central problem in AI: if we can't build LLMs that learn end-to-end from bytes, how will we handle the rest of the world's data that we understand even less? Last year we introduced hierarchical networks (H-Nets) as a first step. Our new version finds up to 3-5x efficiency gains in controlled scaling law experiments, and at larger scale matches an open-source model trained on 11T tokens with ~10x less data and compute. The posts below walk through what we found and why we think it works.

another brutal reality check for the open-sauce china shills GPT-5.5 released on April 23 outscores all tested chinese models and OpenAI and Anthropic have been holding back their strongest models since the supply chain risk designation, so we aren't even seeing the full extent of the domination

the top 20 models on FrontierCode 2 it's actually hilarious it's all OpenAI and Anthropic + Grok 4.7 xhigh at #20# DeepSeek V4.1 Flash 1017 max is rank 52 but yeah totally, china is catching up and we are all doomed