/
Navigation
C
Chronicles
Browse all articles
C
E
Explore
Semantic exploration
E
R
Research
Entity momentum
R
N
Nexus
Correlations & relationships
N
~
Story Arc
Topic evolution
S
Drift Map
Semantic trajectory animation
D
P
Posts
Analysis & commentary
P
Browse
@
Entities
Companies, people, products, technologies
Domains
Browse by publication source
Handles
Browse by social media handle
Detection
?
Concept Search
Semantic similarity search
!
High Impact Stories
Top coverage by position
+
Sentiment Analysis
Positive/negative coverage
*
Anomaly Detection
Unusual coverage patterns
Analysis
vs
Rivalry Report
Compare two entities head-to-head
/\
Semantic Pivots
Narrative discontinuities
!!
Crisis Response
Event recovery patterns
Connected
Nav: C E R N
Search: /
Command: ⌘K
Embeddings: large
VOICE ARCHIVE

Casper Hansen

@casper_hansen_
25 posts
2026-02-25
Is Meta buying more AMD than Nvidia at this point? I wonder if the new 400 series will be comparable to Vera Rubin
2026-02-25 View on X
Wall Street Journal

Meta agrees to acquire up to 6GW of AMD Instinct GPUs in a deal valued at $100B+ that could see Meta own up to 10% of AMD; Meta plans to deploy 1GW in 2026

2026-02-24
Is Meta buying more AMD than Nvidia at this point? I wonder if the new 400 series will be comparable to Vera Rubin
2026-02-24 View on X
Wall Street Journal

Meta agrees to acquire up to 6GW of AMD Instinct GPUs in a deal valued at $100B+ that could see Meta own up to 10% of AMD; Meta plans to deploy 1GW in 2026

The deal could result in Meta owning as much as 10% of AMD's stock as the chip maker seeks to challenge Nvidia

2026-01-27
1T parameter model and multimodal!! Honestly insane how much Kimi is pushing forward the frontier Weights are also on Huggingface, released with INT4 quantization
2026-01-27 View on X
Kimi

Moonshot says Kimi K2.5 builds on K2 with “pretraining over ~15T mixed visual and text tokens” and “can self-direct an agent swarm with up to 100 sub-agents”

Today, we are introducing Kimi K2.5, the most powerful open-source model to date.

1T parameter model released by Qwen!! They finally compare to the best models available out there unlike the rest of open-weight providers Although it's released, unfortunately it's not open weights on Huggingface
2026-01-27 View on X
Qwen

Qwen releases Qwen3-Max-Thinking, its flagship reasoning model that it says demonstrates performance comparable to models such as GPT-5.2 Thinking and Opus 4.5

· QwenTeam丨Translations:.体中文  —  Introduction#  —  We present Qwen3-Max-Thinking, our latest flagship reasoning model.

1T parameter model and multimodal!! Honestly insane how much Kimi is pushing forward the frontier Weights are also on Huggingface, released with INT4 quantization
2026-01-27 View on X
Bloomberg

Chinese startup Moonshot releases Kimi K2.5, saying the model can process text, images, and videos simultaneously and beats its open-source peers in some tests

Alibaba Group Holding Ltd.-backed Moonshot AI released an upgrade of its flagship model, heating up a domestic arms race ahead …

2026-01-26
1T parameter model released by Qwen!! They finally compare to the best models available out there unlike the rest of open-weight providers Although it's released, unfortunately it's not open weights on Huggingface
2026-01-26 View on X
Qwen

Qwen releases Qwen3-Max-Thinking, its flagship reasoning model that it says demonstrates performance comparable to models such as GPT-5.2 Thinking and Opus 4.5

2025-11-29
GLM 4.6 Air might be delayed after this release! Incredible infra work and good improvements on evals
2025-11-29 View on X
Prime Intellect

Prime Intellect debuts INTELLECT-3, an RL-trained 106B parameter open source MOE model it claims outperforms larger models across math, code, science, reasoning

Today, we release INTELLECT-3, a 100B+ parameter Mixture-of-Experts model trained on our RL stack, achieving state …

2025-11-07
K2 Thinking released with Heavy Mode! K2 Thinking Heavy Mode employs an efficient parallel strategy: it first rolls out eight trajectories simultaneously, then reflectively aggregates all outputs to generate the final result. BETTER than gpt-5-pro at HLE :) [image]
2025-11-07 View on X
CNBC

Chinese startup Moonshot releases Kimi K2 Thinking, an open-weight model it claims beats GPT-5 in agentic capabilities; source: the model cost $4.6M to train

Chinese startup Moonshot on Thursday released its latest generative artificial intelligence model which claims to beat OpenAI's ChatGPT in …

2025-10-21
NEW DeepSeek OCR model that outperforms dots ocr while prefilling 3x less tokens [image]
2025-10-21 View on X
The Decoder

DeepSeek releases DeepSeek-OCR, a vision language model designed for efficient vision-text compression, enabling longer contexts with less compute

the new frontier of OCR from @deepseek_ai , exploring optical context compression for LLMs, is running blazingly fast on vLLM ⚡ (~2500 tokens/s on A100-40G) — powered by vllm==0.8....

2025-07-29
GLM 4.5 is 50% cheaper on their Mainland China AI platform until September 1st (called bigmodel). Their GLM 4.5 Air model in FP8 is also entirely free! [image]
2025-07-29 View on X
CNBC

Z.ai, formerly known as Zhipu and that has raised $1.5B from Tencent and others, releases GLM-4.5, an open-source AI model that it says is cheaper than DeepSeek

chinese models really are taking over huh Simon Willison / @simonwillison.net : Pretty decent pelicans from the new GLM-4.5 and GLM-4.5 Air models.  Both models are MIT licensed, r...

o3 competitor: GLM 4.5 by Zhipu AI - hybrid reasoning model (on by default) - trained on 15T tokens - 128k context, 96k output tokens - $0.11 / 1M tokens - MoE: 355B A32B and 106B A12B Benchmark details: - tool calling: 90.6% success rate vs Sonnet's 89.5% vs Kimi K2 86.2% - [image]
2025-07-29 View on X
CNBC

Z.ai, formerly known as Zhipu and that has raised $1.5B from Tencent and others, releases GLM-4.5, an open-source AI model that it says is cheaper than DeepSeek

chinese models really are taking over huh Simon Willison / @simonwillison.net : Pretty decent pelicans from the new GLM-4.5 and GLM-4.5 Air models.  Both models are MIT licensed, r...

2025-05-15
Today I learned Microsoft is retiring the Bing API. Removing it completely. Why? It's super powerful when used as training data for LLMs - look at o3
2025-05-15 View on X
Wired

Microsoft plans to shut down its Bing Search APIs on August 11; a source says the largest customers will retain access, and DuckDuckGo says it won't be affected

Microsoft is limiting access to tools that boosted its rivals, but larger customers like DuckDuckGo say they won't be affected.

How will Microsoft effectively own OpenAI? By dangling Bing API access as a carrot... the only one's to get it, but can quickly be pulled back
2025-05-15 View on X
Wired

Microsoft plans to shut down its Bing Search APIs on August 11; a source says the largest customers will retain access, and DuckDuckGo says it won't be affected

Microsoft is limiting access to tools that boosted its rivals, but larger customers like DuckDuckGo say they won't be affected.

TLDR; Microsoft discovered a gold mine in Bing and now wants it for themself. OpenAI is affected by this too, they can't cut ties with Microsoft now even if they wanted to - unless they work with Google.
2025-05-15 View on X
Wired

Microsoft plans to shut down its Bing Search APIs on August 11; a source says the largest customers will retain access, and DuckDuckGo says it won't be affected

Microsoft is limiting access to tools that boosted its rivals, but larger customers like DuckDuckGo say they won't be affected.

2025-04-08
Meta finetuned a model specifically for LM Arena and didn't tell them?!😱 - Meta should have made it clearer that “Llama-4-Maverick-03-26-Experimental” was a customized model to optimize for human preferences
2025-04-08 View on X
The Verge

LMArena says it is updating its leaderboard policies after a Llama 4 Maverick version, which Meta said in fine print is not public, secured the number two spot

With Llama 4, Meta fudged benchmarks to appear as though its new AI model is better than the competition.

2025-03-12
I often say most code editors today are slot machines. They have gamified it, hacked your brain like the early days of algorithms in social networks
2025-03-12 View on X
The Information

Source: Anthropic's annualized revenue grew from $1B at the end of 2024 to $1.4B in early March; Manus uses tools including Claude 3.7 Sonnet to power its agent

The Information :

2025-02-28
GPT 4.5 pricing is unhinged. If this doesn't have enormous models smell, I will be disappointed [image]
2025-02-28 View on X
TechCrunch

Sam Altman says OpenAI was forced to stagger GPT-4.5's rollout because it is “out of GPUs”; the model is wildly expensive, costing $75 per million input tokens

Hopefully the output is worth it?  🤔  —  Oh... 😥  —  www.theverge.com/news/620021/ ...  [embedded post] X: Ed Zitron / @edzitron : Also, $1.30 per hour per GPU is the Microsoft dis...

GPT 4.5 pricing is unhinged. If this doesn't have enormous models smell, I will be disappointed [image]
2025-02-28 View on X
The Verge

OpenAI launches GPT-4.5, its “most knowledgeable model yet” in research preview, initially warning it's not a frontier model and may perform below o1 or o3-mini

OpenAI's newest and largest model is being released as a research preview.

2025-02-25
Anthropic just dropped Claude Code—a real terminal app, no fluff with 70% performance on SWE Bench. No steep learning curve, unlike Aider. [video]
2025-02-25 View on X
TechCrunch

Anthropic releases Claude 3.7 Sonnet, a hybrid model that can produce fast responses or extended, step-by-step thinking, and Claude Code, an agentic coding tool

and it could be a game changer Ghacks : Anthropic Unveils Claude 3.7: First Hybrid Reasoning AI Model Rowan Cheung / The Rundown AI : Claude enters the reasoning era Siddharth Jind...

Anthropic just dropped Claude Code—a real terminal app, no fluff with 70% performance on SWE Bench. No steep learning curve, unlike Aider. [video]
2025-02-25 View on X
One Useful Thing

Claude 3.7 and Grok-3 are the first “Gen3” models with big gains in handling complex tasks, using 10x more compute than GPT-4-class models, and better reasoning

Note: After publishing this piece, I was contacted by Anthropic who told me that Sonnet 3.7 would not be considered …