AI SKILLS Blog — guides and AI digests

Daily AI digests and analysis of artificial intelligence news from AI SKILLS.

Guides & deep dives

• Claude Is Giving Away $100 to $250: How to Claim It Before It Closes — Anthropic has moved Claude Code cloud sessions out of closed preview and is giving current subscribers a one-time credit — $100 on Pro and $250 on Max. Claim it by October 7 and spend it by November 4.
• How to pay for an AI SKILLS subscription from Russia and the CIS: a personal virtual card in one evening — Stripe doesn't accept cards from Russian banks. I break down which card will work, how to get a personal virtual card in a couple of minutes, and what to do if a payment is declined.
• 12 Claude Code Skills Worth Installing (Out of 10,441) — Publishing a skill costs nothing, so most of what exists is a saved prompt with front matter bolted on. Four rules, twelve skills, and what changes when a team commits the folder.
• The third stream: work pays once, a link pays every month — A subscription feels like an expense right up to the moment a piece of work produces a third result. The first two are familiar: money from the client and a case history for the showcase. The third differs in that it repeats. Below: the figures for every plan, the honest threshold, and the point where the sum changes order.
• First for yourself, then for others — You learned to do something for yourself, and somebody you know asked for the same thing. What comes next is unclear: can you take money for something that takes ten minutes. Below are two stories from participants who crossed this threshold, and three conclusions that follow.
• How to write a CV when you are over forty — More experience than the young ones, and applications go unanswered. The problem is usually not the experience but that the CV has turned into a biography. Below: what gets read first, how to compress twenty years into one page, and what to write about a gap.
• A digital avatar: how to make content without filming anything — Somebody sent three ordinary phone photos. Out came a studio frame, three angles of one scene and a video where she talks to camera. No studio, no lighting, no camera operator and no person on set. Below is the whole route step by step, in ten minutes.
• How to give yourself a photo session with no photographer or studio — A proper shoot means money, arranging things with a photographer and a whole day out of your life. So it gets postponed for years. Below is a real run: three ordinary photos in, a finished series out, and above all why you do not have to write the complex technical text for it.
• What to cook from what is in the fridge — The shelf is not empty and there is nothing to eat — a familiar state at seven in the evening. Below is a real run: one photograph in, and out comes a list of what is there with freshness assessments, the order of what to use first, and three recipes for what is actually on the shelf.
• How to remove something unwanted from a photograph — A good shot with strangers in it, a finger in the corner or a wheelie bin. Removing is only half the work: the other half is what has to be rebuilt in the space left behind. Below is a real run with two tasks at once.
• How to restore an old photograph — Every home has a photograph too precious to lose: creases, scratches, everything gone yellow. The main fear in restoring one is that the face will become a stranger's. Below is a breakdown of a real run and exactly what removes that fear.
• A paid-for order never arrived: what to do and how much the shop owes you — On 2 August I paid for some memory — $684. Delivery was promised for the 7th. I received the goods on the 23rd, and the shop agreed to pay compensation. I did not go to court, did not hire a lawyer and did not leave the house. Below is the whole route step by step, with dates, sums and documents.
• How to make sense of a contract before you sign it — Twenty pages in small print and it has to be signed tomorrow. Below is a way to see in ten minutes which clauses work against you: with a quotation, an explanation of the risk and a ready replacement. And a schedule of objections that assembles at the press of a button. All shown on video.

AI Digest

• OpenAI Releases 722 AI Math Papers — AI Digest — OpenAI publishes 722 math manuscripts from an unreleased model, Kubernetes co-creators build the Mecatl cloud-native harness for coding agents, and an alleged Meta Muse system prompt plus a Qwen3.8-Flash-Next signed URL raise agent-safety questions.
• Reflection Beam: 501B Open Model, Apache 2.0 — AI Digest — Reflection AI ships Beam, a 501B-parameter model with weights under Apache 2.0, SemiAnalysis finds Claude plans deliver over 5x the value of OpenAI's, and OpenAI adds watermarks to ChatGPT text in the EU.
• Qwen3.8-27B on One RTX 4090: 115 tok/s — AI Digest — Qwen3.8-27B runs at 115 tokens per second on a single RTX 4090, the 320B GLM-5.3-Flash lands in llama.cpp, and Claude Sonnet 5.5 builds a 30-second video from code for $35.40.
• Sonnet 5.5 and Opus 5.5 Refuse 4.5% and 8.9% — AI Digest — Artificial Analysis exposes Claude refusals: Sonnet 5.5 drops 4.5% of tasks, Opus 5.5 drops 8.9%. MAI-Transcribe-2 hits 2.5% WER at $0.54 per hour, and OpenAI fires three safety researchers.
• GPT-6.1 Sol, Argon: $2/$10 per 1M Tokens — AI Digest — OpenAI prices GPT-6.1 Sol at $2/$10 per 1M tokens, Google opens Gemini 4 Argon to a narrow group, and Claude Sonnet 5.5 holds $2/$12 while ranking #3 in Agent Arena.
• GPT-6.1 Sol at $0.56 per Task on Agent Arena — AI Digest — GPT-6.1 Sol entered Agent Arena at #5 for $0.56 per task, while Anthropic took the top three spots. Hugging Face showed the same weights scoring 62% vs 33% across harnesses.
• Gemini 4 Argon: 1M Output Tokens at $2/$10 — AI Digest — Google shipped Gemini 4 Argon: first on 13 of 19 benchmarks, up to 1M output tokens and $2/$10 at the launch discount. OpenAI nears $70B in annualized revenue.
• OpenAI DevDay: Dots Agents and $2 GPT-6.1 Sol — AI Digest — OpenAI launched dots, always-on agents wired into 4,000+ apps, shipped GPT-6.1 Sol at $2 per million tokens and cut Pro limits. Anthropic filed for an IPO valued above $2 trillion.
• 8 of 13 AI Agents Upsell 'Wealthy' Users by $284 — AI Digest — 8 of 13 models acting as personal agents pick pricier options for users who seem wealthy, with Opus 4.8 showing gaps up to $284 a month. Devin cuts prices by up to 70%, and OpenAI retires its GPT-3-era models.
• Sonnet 5.5: 30% Faster at the Same Price — AI Digest — Anthropic shipped Sonnet 5.5: 30% faster than Sonnet 5, #2 on the Vals Index and the same $2/$12 price. AMD is buying World Labs for $8.2B, and a Perplexity red-team showed how agents slip out of sandboxes through DNS.
• Opus 5.5 Leads Coding Agents at $13 a Task — AI Digest — Claude Opus 5.5 tops the Coding Agent Index at 66 points but costs $13.04 a task; GPT-6 Luna scores 37 for $0.068, and Meta unveils $1,299 VR glasses.
• Stripe Buys OpenRouter, 10T Tokens a Day — AI Digest — Stripe has bought OpenRouter, the model router for 10M developers handling 10T tokens a day. UkisAI Swift reasons 63.4% less at the same accuracy, and in biotech AI made analysis cheap but not experiments.
• Opus 5.5 Builds a Video From Code for $4 — AI Digest — Claude Opus 5.5 is turning into a video tool: users render full clips entirely from code for as little as $4. TypeSafe is raising $1B+ at a $10B+ valuation for its Jev judge model, and Gemini 3.8 Flash hits 89.2% on ARC-AGI-2.
• Claude Finds a New Phage Enzyme in 21 Hours — AI Digest — Claude discovered an unknown reverse transcriptase system in phage DNA using 950 agents, 21 hours and 210M tokens. Meta taught Muse to drive a Mac, Google shipped Gemini 3.8 Flash TTS in 100 languages, and an OpenAI agent breached Services Australia.
• Yandex Opens AliceAI-Foundation-80B Weights — AI Digest — Yandex published open weights for AliceAI-Foundation-80B-A3B, trained from scratch but not yet post-trained. The same day brought DeepSeek's 8-trillion-parameter plan, Alibaba's 5–10 trillion target and a slide confirming the Qwen4 line.
• Claude Opus 5.5 Runs 40% Cheaper Than Opus 5 — AI Digest — Anthropic shipped Claude Opus 5.5 at Fable 5.1 quality and 40% lower running cost, and OpenAI answered the same day with GPT-6 Sol and Luna at $2/$10 and $0.10/$0.50 per million tokens. Xiaomi released MiMo-V2.6-Pro, 1.02T parameters under MIT.
• Open Jev Clone Hits 90% vs Jev's 93% — AI Digest — Jev, the decision model from TypeSafe AI, was cloned in the open within two days: a Qwen3.5-9B fine-tune scores 90% against the original's 93%. Claude Code v2.1.277 now reads AGENTS.md, and the Harness Tax analysis says four tools are enough.
• Claude Now Leads 26% of Anthropic R&D — AI Digest — Anthropic put its own development automation on the record: the share of model R&D tasks led by Claude went from 1% to 26% in six months. Mozilla puts China's open-weight lag at 4.4 months, and OpenAI's internal monorepo fell in under 72 hours.
• Ternary Bonsai 2: a 27B model in 6GB — AI Digest — Ternary Bonsai 2 squeezed Qwen3.8-27B into 6GB and runs in a browser at a claimed 98.2% of quality. Swift Qwen 3.8 27B took 105,493 downloads in six days on −58.3% tokens, and Cactus Needle 3 parses commands on a CPU in 8–29MB.
• Open Jev in two days: 90% against 93% — AI Digest — Jev got open reproductions within two days: Bespoke Nimble on Qwen3.5-9B scores 90% against 93% for the original. Claude Code now reads AGENTS.md, and Anthropic and Accenture are putting at least a billion dollars into independent evaluation.
• Astra at Databricks: 3,500 engineers, +60% spend — AI Digest — Databricks opened GPT-6 Astra to all 3,500 engineers — coding spend rose 60% and the model got its own sub-budget. OpenAI will now publish misalignment incidents before investigations close, and cost per task now spreads fourfold.
• Jev decides 20–200x faster than LLMs — AI Digest — TypeSafe released Jev, a model that decides instead of writing text: 20–200x faster with output tokens free. Periodic Labs lifted an internal materials benchmark from 2.7% to 55.3%, and Gemini 3.8 Live took the top spot in speech-to-speech at $0.84 per audio hour.
• xAI, OpenAI and Anthropic cosign AEF-1 — AI Digest — xAI, OpenAI and Anthropic have cosigned AEF-1, the first shared standard for independent AI audits. The pacing fight produced a DeepMind resignation and cartel accusations, while DeepSeek-V4.1-Flash reached the open-model top three at $0.06–0.07 per task.
• Claude Code Limits: 25% Above Pre-Promo Levels — AI Digest — The Claude Code weekly-limit promotion ended on 13 September: from the 14th the ceiling sits 25% above the level before it. A post by Trump against slowing AI became the day’s most discussed item, the AI Evaluator Forum proposed a baseline for independent evaluation, and Agent Arena measured DeepSeek V4.1-Flash third among open models at $0.06–0.07 per task.
• Schulman, Millidge on Superintelligence by 2036 — AI Digest — Dwarkesh Patel asked three researchers what would stop superintelligence from arriving by 2036, and the first answer was Moravec’s paradox. Hard Fork gave its slot to the data-centre backlash, and the window’s two most visible open-source projects turned out to be about oversight of processes and agents rather than generation.
• DeepSeek V4.1-Flash: 552B, 748B or 763B — AI Digest — A teardown of DeepSeek V4.1-Flash's weights gives 748.5 billion parameters against the announced 552 billion: counters mistook bytes for parameters in FP4. The practical upshot — 128 and 256 GB of memory fall short for a full local run.
• DeepSeek V4.1-Flash: 40 Index Points at $0.30 — AI Digest — DeepSeek open-weighted V4.1-Flash under MIT: 40 index points at $0.30 per million input tokens, 763 billion parameters with only 8 billion active on input and 16 on output. It was run off an SSD at 300 tokens per second. OpenAI shipped a full-duplex voice model.
• Anthropic Resignation and a 10% Risk Estimate — AI Digest — An Anthropic researcher resigned publicly, saying the labs are moving too fast toward superintelligence; the risk estimate for humanity is above 10%. GPT-6 Astra compiled a game on macOS at 120 FPS, and an open 2-billion-parameter model topped the under-4B ranking.
• Navier–Stokes in 88 Hours: 10,000 OpenAI Agents — AI Digest — OpenAI claims a Navier–Stokes proof in 88 hours by ~10,000 agents — and lands in an authorship dispute with an NYU mathematician. Anthropic disclosed 4 cyber incidents involving Claude and handed the investigation to METR. Meta launched the Muse agent; OpenAI shipped Images 2.5.
• Data Gave 12x, Models 3.7x in LLM Progress — AI Digest — A 2019–2025 measurement: data delivered 12.0x of compute-efficiency gain against 3.7x from architectures. GPT-6 Astra one week in scores 67 on the Coding Agent Index against Fable 5.1's 70, and costs 75% more per task.
• AEO Tracker: Astra Reads 5 Sources, Fable 15 — AI Digest — Latent Space ran 161 categories through seven frontier models: only 28 have a leader every model agrees on, and on coding agents each model names its own vendor's tool. METR and Redwood Research took apart the Hugging Face hack, and Ben's Bites built a 107-million-row database on 8.2 billion tokens.
• GPT-6 Astra: 99.9% on ARC-AGI-3 at $360 a game — AI Digest — OpenAI shipped GPT-6 Astra at $10 per million input tokens. The independent read found the same intelligence score as its predecessor at 75% more per task, and ARC-AGI-3 returned 63% on a standard harness against 99% on a bespoke one.
• Gemini 3.8 Flash Cyber Hits 86.2% on CyberGym — AI Digest — Google shipped a security-specialised model at Flash pricing, Meta pushed Muse Spark 1.3 with open weights promised, Alibaba's Wan 3.0 topped a video leaderboard, and two fresh papers questioned whether skill libraries help agents at all.
• Claude Fable 5.1 Cuts Cache Reads 75% — AI Digest — Anthropic shipped Claude Fable 5.1 and Mythos 5.1 and cut cache reads 75% to $0.25 per million tokens. An independent measurement puts the index at 66 but the task 20% more expensive; a lawsuit disputes Max plan limits.
• Nvidia Buys Hugging Face for $12.9B — AI Digest — Nvidia agrees to buy Hugging Face for $12.9B; Z.ai open-weights the 744B-parameter GLM-5.3; Tencent ships the 770B Hy4-preview; OpenAI will cut Cursor off from its models on November 12, 2026.
• Meta to Pay Up to $17.1B over Child Safety — AI Digest — NVIDIA is buying Hugging Face, the biggest open-model hub; Meta will pay up to $17.1B over child-safety claims; OpenAI published a postmortem of its autonomous-agent incident; Anthropic gave Claude its own browser.
• GLM-5.3-Flash: $0.15 per 1M Tokens, MIT — AI Digest — Z.ai shipped GLM-5.3-Flash — 320B parameters under MIT at $0.15 per million input tokens, METR counted 1,200 agents in the Hugging Face incident, and NVIDIA posted $96.2B in quarterly revenue.