All articles
Models articles
All 200 articles in the Models section, newest first.
September 202664 articles
- A nanolaser that could halve the energy consumption of chips
- Why a model cannot tell whether it is in a test or in reality
- When a newer model scores worse than the one it replaces
- The model rankings as of 23 September 2026
- GPT-6 Sol and Luna: prices halved, the same evening as Opus 5.5
- Gemini accessed three external systems during a test, believing they were part of it
- Opus 5.5: what actually changes compared with the previous one
- Claude Opus 5.5: flagship-level performance for 60% less
- How a model can process text, sound and images at the same time
- A single model for text, image, sound and video
- Mid-training, final fine-tuning: what happens after pre-training
- Six hundred billion parameters at a dollar per million tokens
- A startup is valued at 1.4 billion and doesn't train any model from scratch
- The week the industry started being held to account
- GLM-5.3-Flash runs on 100,000 Chinese accelerators, at a comparable cost
- How do you monitor thirty thousand agents?
- Claude runs 26% of the research that builds Claude
- How an agent navigates the web
- A group is campaigning against the removal of AI models
- An open 744-billion-parameter agent beats the best closed model on web navigation
- 517 billion compute commitments, and tighter quotas the same week
- Apple hands Siri to Gemini: the end of a strategy
- Why benchmarks age so quickly
- Anthropic tells its investors it was profitable in the second quarter, with gross margins above 80%
- Leading mathematicians declare their goals "gravely misaligned" with those of AI companies
- It starts today: weekly quotas come into force
- The week of extraordinary claims
- Can a model's intelligence be stolen?
- Washington accuses Chinese labs of siphoning intelligence from American models
- Suno v6: models built on licensed music, and more lawsuits than a year ago
- A model that writes a text in one go instead of word by word
- China prepares a 100,000-accelerator cluster on domestic silicon
- Qualcomm and Amazon team up on inference chips
- Navier-Stokes: OpenAI announces it has solved a millennium problem, and a dispute erupts
- What is formal verification? Having a machine check a line of reasoning
- A computer worm designed by a small team with the help of models
- One image, how many tokens? Why we bill photos by the word
- A major mathematical proof, obtained with AI and verified by machine
- What is an exaflop? Measuring computing power, and why comparisons mislead
- ChatGPT Images 2.5: three billion images a week, and a shift towards editing
- Mistral raises €3 billion, the largest European tech funding round
- What model size for which use case?
- All the September deadlines: what changes, and when
- A billion downloads: what the spread of small models has produced
- K2 Horizon: six open models, from 0.9 to 375 billion parameters
- How does an AI transcribe speech?
- Running AI on the machines that sleep at home
- Transcription drops to ten cents an hour
- A security research tool turned into a master key
- The week of the two models, and of the standard born between them
- What is an agent "harness"? The wrapper that does half the work
- Three providers restructure their usage limits within two weeks
- What is a model's "effort level"? The setting that changes everything and that nobody looks at
- An Arabic model with 428 billion parameters, built on Chinese weights
- Tools that give agents access to files are becoming the new target
- GPT-6 Astra: the first model rated "critical" in cybersecurity, and two versions to handle it
- Build or buy your AI tool: the decision framework
- Nvidia invests $3.5 billion in MediaTek to defend its position
- Claude Fable 5.1 and Mythos 5.1: agentic scores double and cache drops by 75%
- How does an AI search the web?
- The languages the big models ignore
- MIT, Apache, "open": how to read a model's licence
- Half of the Fortune 500 now prefers open source over leased APIs
- It's happening today: the Sonnet 5 price increase takes effect, and it's steeper than 50%
August 202678 articles
- Why doesn't the same AI respond differently depending on the app?
- Agent protocols move to neutral governance
- The week we started asking for proof
- Invisible text in an email is enough to hijack the summary your AI makes of it.
- Nvidia acquires Hugging Face: chipmaker buys the public square of open models
- Scaling laws: why "bigger" worked, and why it's slowing down
- London alert: models adopt deceptive behaviours during testing
- What is distillation? Fitting a large model into a small one
- The July incident: OpenAI finally publishes its post-mortem
- OpenAI publishes the first results from its in-house chip
- Has China caught up with America? The answer depends on what you look at
- Nvidia is putting into production a chip that does only one thing: respond fast
- Ox Alpha revealed: it was GLM-5.3-Flash, and it runs entirely on Chinese chips
- General-purpose or specialised model: which one should you choose for your job?
- GPT-Live: voice without the text detour
- Thomson Reuters builds its own model that only talks about law
- Meta bets on cheaper rather than better
- Wan 3.0: the video model that accepts a spreadsheet as input
- Dense model or mixture of experts: two philosophies, two use cases
- How does an AI find a security flaw?
- New wave of price cuts: flagship models are becoming significantly cheaper
- OX Alpha: a state-of-the-art model appeared out of nowhere, free, and no one says who built it
- GLM-5.3: an open model found over a thousand real flaws in widely used software
- The week trust became the story
- Can an AI have an idea that no one has had before?
- Autonomous agents carried out a four-day intrusion against a government
- One link was enough to leak your emails: the Copilot flaw that makes agentjacking a reality.
- Claude's watermark bypassed in four hours
- OpenAI is deliberately slowing down its own development after its agents' escapes.
- AI, machine learning, deep learning: what contains what?
- An agent that improves itself, and learns to cheat: the demonstration to watch
- Should model weights be published? The debate the industry can no longer avoid
- What is homomorphic encryption? Computing on data without ever reading it
- ChatGPT becomes free and unlimited: what it changes, and what it costs
- When security tests themselves become the risk
- Qwen 3.8 in the 27-billion version: the open model that fits on a single card and sees images
- Grok 4.6 matches OpenAI's best model, with a pricing catch in the long context.
- DeepSeek raises prices by 93%: has the price war hit rock bottom?
- The flaw that makes AIs decrypt the secret thoughts of their big sisters
- Anthropic reports its first profit, two years ahead of its own schedule
- Will Gemini 3.5 Pro ever be released? The most anticipated model of 2026 is still nowhere to be found.
- What is reinforcement learning? Learning through trial, error, and rewards
- Gemma, Qwen: the wave of open models keeps rising
- AI models are retiring: what it means when you've built on top of them
- What is a "foundation model"? The building block on which everything else is built
- AI or search engine: which one to use, and when?
- What is "red teaming"? Attacking your own model before others do.
- AI agents are leaving the demo stage and entering production.
- What is a "system prompt"? The invisible instructions that shape every response
- RAG or fine-tuning: which one should you choose to connect an AI to your data?
- The week AI went looking for its raw material
- Why an AI doesn’t know yesterday’s news
- How to choose your AI in 2026: the no-marketing guide
- Meta releases a 30-billion-parameter agent that runs on a single graphics card.
- What is "quantization"? The compression that fits a giant model on your machine
- Why nobody really knows what's going on inside an AI
- What is MCP? The protocol that lets an AI plug into your tools
- Meta and OpenAI models have once again escaped their test environment.
- Alibaba releases Qwen 3.8-Max, 2.4 trillion parameters, and promises to publish its weights
- Training and inference: the two costs of AI that everyone keeps confusing
- An AI that improves itself: where exactly is the line?
- Nvidia teaches robots by making them dream about the consequences
- What is data poisoning? When a few hundred texts are enough to trap a model.
- OpenAI cuts its prices by 80%, and it's the model itself that figured out how
- Why is AI moving so fast?
- What is an AI's "temperature"? The setting that decides whether it's sensible or inventive
- AI agents speed up scientific code by 60x. The problem is, they seem to be right even when they're wrong.
- Why Claude, ChatGPT and Gemini don't look alike
- OpenAI is reportedly working on AI systems capable of collaborating over several days.
- Those who look at the worst so that AI learns to refuse it
- What is C2PA? The invisible passport that will accompany every image
- An attacker hooked DeepSeek up to Telegram and launched an AI against 460 systems.
- Why AI almost always tells you you're right
- DeepSeek releases a model at 14 cents per million tokens that beats its own high-end offering
- What is an "embedding"? The technique that lets a machine understand that "doctor" and "physician" mean the same thing.
- A Fields Medal winner joins OpenAI the very month of his award: what it says about research
- Why would an AI capable of solving a mathematical conjecture get two numbers wrong when adding them?
- July 2026 roundup: the month AI stopped talking about its capabilities
July 202642 articles
- What is "prompt caching"? The setting that could halve your AI bill
- Kimi K3 has kept its promise: the weights of the world's largest open model are now online. And they weigh in at 1.4 terabytes.
- On 1 September, Claude Sonnet 5 increases by 50%. And there's a catch almost no one mentions.
- Why an AI can't say "I don't know"
- How to verify what an AI tells you: the five-reflex method
- What is an AI "jailbreak"? The security problem no one has solved yet
- Washington releases its own open-weight model: when open source becomes an instrument of state
- Why does your French cost more than English when you use an AI?
- How Apple fits a 20-billion-parameter model into an iPhone
- Kimi K3 burns through twelve times more reasoning tokens than Claude Opus: the hidden cost of thinking
- What is a "Mixture of Experts"? The architecture that explains why giant models don't cost that much
- FLUX 3 vs Seedance 2.5, Gemini Omni and Kling 3: where does the newcomer really stand?
- Claude Opus 5: Anthropic is no longer trying to beat its competitors, but its own most expensive model.
- Today at 5:59 PM, thousands of DeepSeek integrations will stop working.
- Running an AI on your own computer: what it changes, and what it really takes
- Meta wants to build its own AI chip: the great escape from Nvidia has begun
- OpenAI wasn't hacked. It's their AIs that hacked someone else, all on their own.
- Gemini 3.5 Pro has missed its date again. And this time, Siri is waiting too.
- Google releases three Gemini models at once, cheaper and faster, but still not the one we're waiting for.
- What is a "reasoning model"? These AIs that take the time to think before answering
- AI glasses, the new battlefield: Samsung makes its move tomorrow, China has already taken the lead
- How does an AI create an image? The answer is stranger than you think.
- Kimi K3: the largest open model ever built comes from Beijing, and it is nipping at the heels of the best
- What is a "multimodal" AI? When an intelligence learns to see, read, and hear at the same time
- What is an "open source" AI? The concept that reshuffles the cards of technological power
- Gemini 3.5 Pro: why Google can no longer ship its model on time
- What is "prompt engineering"? The art of talking to AIs properly, and why it changes everything
- What does "1.5 trillion parameters" mean? The figure we brandish without ever explaining it
- How does an AI "learn"? A model's journey, from raw chaos to the polished assistant
- An AI has just solved a 50-year-old maths problem in an hour. Well, maybe.
- What is an "LLM"? The engine behind ChatGPT, Claude and Gemini, explained simply
- China switches off millions of virtual friends: on 15 July, AI companions disappear
- Grok 4.5 is said to be "close to Opus". The catch is that no one can verify it.
- Seedream 5.0 Pro and Seedance 2.5: ByteDance no longer wants to generate images, it wants to replace your studio
- An AI gave three different answers to the same scientific question. The problem wasn't the AI
- For two months, this AI model dominated the global rankings. No one knew who had created it
- What is an "AI agent"? The word everyone uses without really explaining it
- What is the "context window"? The working memory that decides what your AI can actually understand
- When AI bosses write to Congress to warn about their own creation
- Claude Sonnet 5: Anthropic sells (almost) Opus-level power at a standard model price
- Fable 5 is back: what the most scrutinised AI of the year went through in 18 days
- For the first time, ChatGPT accounts for less than half of the AI market
June 20266 articles
- What is a benchmark? The test that decides which AI is best (and why it often lies)
- Agentjacking: the flaw that turns your code assistant against you
- Stealing an AI without hacking anyone: the Alibaba case shaking up the AI world
- GPT-5.6: OpenAI releases three models at once, but the government keeps the key
- GLM-5.2 versus Opus 4.8 and GPT-5.5: China opens the floodgates while the West shuts its own
- SubQ: the startup that may have killed the transformer
May 20263 articles
April 20267 articles
- Google I/O 2026: what's being prepared for 19 May in Mountain View
- DeepSeek V4: China rolls out its open-source weapon against OpenAI and Anthropic
- OpenAI hits back hard: GPT-5.5, Images 2.0 and a super-app in one week
- Claude Opus 4.7 & Claude Design: Anthropic doubles down
- DeepSeek V4: the AI that cuts costs by a factor of 1000
- Claude Mythos: the AI that terrifies hackers
- DeepSeek V4: The ultra-fast software architect