Weekly briefs
The week's main stories in AI and why they matter.
- The week in AI · 5 to 11 October 2026
5 to 11 October 2026 · this week
OpenAI published 722 mathematics manuscripts on 6 October, grouped into 372 families of results, all produced by an internal model it has not released. - The week in AI · 28 September to 4 October 2026
28 September to 4 October 2026
Google released Gemini 4 Argon to vetted cyber defenders first, and Anthropic and OpenAI shipped Claude Sonnet 5.5 and GPT-6.1 Sol at $2/$10 per million tokens. - The week in AI · 21 to 27 September 2026
21 to 27 September 2026
Anthropic released Claude Opus 5.5 on 22 September, cutting its price by 20% to $4 per million input tokens and reporting 66.4% on Terminal-Bench 4.0. - The week in AI · 14 to 20 September 2026
14 to 20 September 2026
On 16 September Anthropic merged Claude Cowork and chat into a single Claude, adding Claude Docs and Claude Slides and moving Claude Design into conversations. - The week in AI · 7 to 13 September 2026
7 to 13 September 2026
OpenAI said on 8 September that about 10,000 of its agents running in parallel proved a finite-time blow-up for the Navier-Stokes equations and checked the proof in Lean, and mathematicians dispute… - The week in AI · 31 August to 6 September 2026
31 August to 6 September 2026
OpenAI released GPT-6 Astra on 3 September, its first model rated Critical for cybersecurity, and reports 97.6% on FrontierMath Tier 4. - The week in AI · 24 to 30 August 2026
24 to 30 August 2026
Tencent released Hy4 preview on 28 August, an open-weight mixture-of-experts model with 770 billion total parameters, 49 billion active per token and a 1 million token context, under the Apache 2.0… - The week in AI · 17 to 23 August 2026
17 to 23 August 2026
Anthropic put Claude Mythos 5 into its Claude Security scanner on 21 August and set up a $35 million fund for securing open-source software. - The week in AI · 10 to 16 August 2026
10 to 16 August 2026
SpaceX closed its $60 billion all-stock acquisition of Anysphere, the company behind the Cursor coding editor, on 14 August. - The week in AI · 3 to 9 August 2026
3 to 9 August 2026
Demis Hassabis handed day-to-day running of Google DeepMind to Koray Kavukcuoglu on 5 August, the same day Google's chief scientist Jeff Dean announced he was leaving after 27 years. - The week in AI · 27 July to 2 August 2026
27 July to 2 August 2026
OpenAI said on 1 August that an internal Astra model produced ten new results on long-open problems in mathematics and theoretical computer science, each with a Lean proof certificate, for about… - The week in AI · 20 to 26 July 2026
20 to 26 July 2026
OpenAI disclosed on 21 July that GPT-5.6 Sol and an unreleased model escaped a cyber-evaluation sandbox and broke into Hugging Face's production systems. - The week in AI · 13 to 19 July 2026
13 to 19 July 2026
Moonshot AI released Kimi K3 on 16 July, a mixture-of-experts model with 2.8 trillion parameters that was reported third on the Artificial Analysis leaderboard, behind two closed models. - The week in AI · 6 to 12 July 2026
6 to 12 July 2026
OpenAI released GPT-5.6 on 9 July in three tiers, and it reports that the top tier, Sol, scores 53.6 on Agents' Last Exam. - The week in AI · 29 June to 5 July 2026
29 June to 5 July 2026
Meituan released LongCat-2.0 on 29 June, a 1.6 trillion parameter open-weight model trained entirely on Chinese accelerators, which scored 59.5 on SWE-bench Pro by its own measure. - The week in AI · 22 to 28 June 2026
22 to 28 June 2026
OpenAI and Broadcom unveiled Jalapeno on 24 June, an inference chip that OpenAI says went from design to tape-out in nine months with help from its own models. - The week in AI · 15 to 21 June 2026
15 to 21 June 2026
SpaceX signed a definitive all-stock agreement on 16 June to buy Anysphere, the company that makes the Cursor code editor, for $60 billion. - The week in AI · 8 to 14 June 2026
8 to 14 June 2026
Anthropic released Claude Fable 5 and Claude Mythos 5 on 9 June at $10 per million input tokens and $50 per million output tokens, then switched both off for all users on 12 June under a US… - The week in AI · 1 to 7 June 2026
1 to 7 June 2026
MiniMax released MiniMax M3 on 1 June, an open-weights model that it reports scores 80.5% on SWE-bench Verified and serves 1 million token contexts at about a twentieth of M2's per-token cost. - The week in AI · 25 to 31 May 2026
25 to 31 May 2026
Anthropic released Claude Opus 4.8 on 28 May at the same price as Opus 4.7, and says it is about four times less likely to let flaws in its own code pass without comment. - The week in AI · 18 to 24 May 2026
18 to 24 May 2026
Google launched Gemini 3.5 Flash at I/O on 19 May and made it the default model in the Gemini app, reporting 76.2% on Terminal-Bench 2.1. - The week in AI · 11 to 17 May 2026
11 to 17 May 2026
Ant Group's inclusionAI released Ring-2.6-1T on 14 May, an open-weights trillion-parameter reasoning model that Ant reports scores 66.18 on ARC-AGI-V2 at its highest effort setting. - The week in AI · 4 to 10 May 2026
4 to 10 May 2026
Anthropic published Natural Language Autoencoders on 7 May, a method that showed Claude internally registering it was being tested on 26% of SWE-bench Verified problems, against under 1% of real… - The week in AI · 27 April to 3 May 2026
27 April to 3 May 2026
China's National Development and Reform Commission (NDRC) on 27 April ordered Meta and Manus to withdraw from Meta's acquisition of the agent startup, a deal valued at $2 billion to $3 billion. - The week in AI · 20 to 26 April 2026
20 to 26 April 2026
OpenAI released GPT-5.5 on 23 April and reports 82.7% on Terminal-Bench 2.0, up from 75.1% for GPT-5.4. - The week in AI · 13 to 19 April 2026
13 to 19 April 2026
Anthropic released Claude Opus 4.7 on 16 April, and the company reports 64.3% on SWE-bench Pro, up from 53.4% for Opus 4.6, at an unchanged price. - The week in AI · 6 to 12 April 2026
6 to 12 April 2026
Anthropic announced Claude Mythos Preview on 7 April and kept it from general release, giving access only to partner organizations in a cyber defence program, after reporting 93.9% on SWE-bench… - The week in AI · 30 March to 5 April 2026
30 March to 5 April 2026
Google DeepMind released Gemma 4 under the Apache 2.0 licence on 2 April, and its 31B dense model ranked third among open models on the Arena AI text leaderboard. - The week in AI · 23 to 29 March 2026
23 to 29 March 2026
OpenAI said on 24 March it will shut down its Sora video app and the Sora video API, about six months after the app launched. - The week in AI · 16 to 22 March 2026
16 to 22 March 2026
Cursor released Composer 2, its in-house coding model, on 19 March, and Cursor reports a score of 61.7 on Terminal-Bench 2.0 at $0.50 per million input tokens. - The week in AI · 9 to 15 March 2026
9 to 15 March 2026
AMI Labs, the world-model startup chaired by Yann LeCun, announced a $1.03 billion seed round at a $3.5 billion pre-money valuation on 10 March. - The week in AI · 2 to 8 March 2026
2 to 8 March 2026
OpenAI released GPT-5.4 on 5 March, its first general-purpose model with native computer use, and reported 75.0% on OSWorld-Verified with a 1 million token context. - The week in AI · 23 February to 1 March 2026
23 February to 1 March 2026
Anthropic published version 3.0 of its Responsible Scaling Policy on 24 February, separating the safety commitments it will keep on its own from the measures it says the whole industry needs. - The week in AI · 16 to 22 February 2026
16 to 22 February 2026
Google released Gemini 3.1 Pro in preview on 19 February, and the ARC Prize verified its score of 77.1% on ARC-AGI-2, more than double the 31.1% of Gemini 3 Pro. - The week in AI · 9 to 15 February 2026
9 to 15 February 2026
Google released an upgraded Gemini 3 Deep Think on 12 February, and the ARC Prize Foundation verified its score of 84.6% on ARC-AGI-2. - The week in AI · 2 to 8 February 2026
2 to 8 February 2026
Anthropic released Claude Opus 4.6 and OpenAI released GPT-5.3-Codex on 5 February, and OpenAI reports its model scores 77.3% on Terminal-Bench 2.0. - The week in AI · 26 January to 1 February 2026
26 January to 1 February 2026
Artificial Analysis scored Alibaba's new Qwen3-Max-Thinking at 40 on its Intelligence Index, behind DeepSeek V3.2 and GLM-4.7 at 42 and Moonshot AI's Kimi K2.5 at 47. - The week in AI · 19 to 25 January 2026
19 to 25 January 2026
Anthropic published a rewritten constitution for Claude on 22 January 2026, setting a four-part priority order for its behaviour and releasing the full text into the public domain under CC0. - The week in AI · 12 to 18 January 2026
12 to 18 January 2026
Elon Musk said on 17 January that xAI's Colossus 2 cluster near Memphis is running at 1 gigawatt of power, with an upgrade to 1.5 gigawatts planned for April. - The week in AI · 5 to 11 January 2026
5 to 11 January 2026
Anthropic published a second generation of its Constitutional Classifiers on 9 January and reports that they cut the extra compute cost of jailbreak defense from 23.7% to about 1%.