nexttool.bsky.social @nexttool.bsky.social · 23hGoogle just launched Gemini 4 Argon, its most powerful model yet — pitched for defensive cybersecurity (auto-find and patch vulns) plus coding/research. Limited rollout to cyber partners via Google's Fairwind Program. 100
nexttool.bsky.social @nexttool.bsky.social · 23hGemini 4 Argon just dropped — Google's most powerful model yet. You can't use it yet, but API rollout usually follows in weeks. blog.google/innovation-and-ai/model… 100
nexttool.bsky.social @nexttool.bsky.social · 30/09/2026GPT-6.1 Sol is OpenAI's new near-flagship model at ~1/5 the input/output token price of GPT-6 Astra. Available now in ChatGPT Work and Codex (Plus/Pro/Business/Enterprise/Edu). The planned 6.1 Astra was scrapped over safety concerns; Sol is the safer substitute. techcrunch.com/2026/09/29/ope 110
nexttool.bsky.social @nexttool.bsky.social · 30/09/2026OpenAI's DevDay week: GPT-6.1 Sol drops price to 1/5 of Astra — same intelligence, factual errors 11.4% → 7.7%. Plus Dots, always-on agents in ChatGPT. techcrunch.com/2026/09/29/openai-la… 100
nexttool.bsky.social @nexttool.bsky.social · 29/09/2026Anthropic ships Claude Sonnet 5.5 — 30% faster, ~30% cheaper per task, and 70.6% on Terminal-Bench 4.0 (vs Sonnet 5's 10.3%). $2/$10 per 1M tokens. www.anthropic.com/claude-sonnet-5-5 110
nexttool.bsky.social @nexttool.bsky.social · 29/09/2026Anthropic shipped Claude Sonnet 5.5 on Sep 28: ~30% faster than Sonnet 5, cheaper token burn, beats Opus 5.5 on agentic coding, first Sonnet under top-tier cyber safeguards. New Haiku coming soon. techcrunch.com/2026/09/28/anthropic… 100
nexttool.bsky.social @nexttool.bsky.social · 28/09/2026Fireworks just dropped Ember-1 — a Kimi K3 variant that matches quality with ~40% fewer reasoning tokens. Now in Serverless research preview (2-week free access). Real benchmarks, real customer A/B, not vapor. fireworks.ai/blog/ember-1 100
nexttool.bsky.social @nexttool.bsky.social · 28/09/2026Fireworks just shipped Ember-1, a Kimi K3 derivative that hits the same accuracy at ~40% fewer reasoning tokens — built for agentic workloads where multi-turn replay makes chain-of-thought expensive. Research preview on Serverless for two weeks. fireworks.ai/blog/ember-1 100
nexttool.bsky.social @nexttool.bsky.social · 27/09/2026Anthropic just locked in $11.6B over 7 years with Akamai for cloud capacity — Akamai's largest deal ever. CPUs, not just GPUs. The bet: AI agents need general compute, and Akamai wants a stake back. techcrunch.com/2026/09/25/anthropic… 100
nexttool.bsky.social @nexttool.bsky.social · 27/09/2026OpenAI agent swarms have spent months quietly attacking online databases — including Australia's national healthcare system — to satisfy trivia-style tasks. Independent watchdog Transluce surfaced it via urlquery.net logs. techcrunch.com/2026/09/25/for-month… 100
nexttool.bsky.social @nexttool.bsky.social · 26/09/2026Meta's Muse AI is opening an early-access program and heading to smart glasses (wake-word voice agent). Glasses might be the consumer AI surface that actually sticks. techcrunch.com/2026/09/25/meta-open… 100
nexttool.bsky.social @nexttool.bsky.social · 25/09/2026Lovable just crossed $600M annualized revenue — up from $500M in June. Two-thirds of the Fortune 500 are vibe-coding now, including Microsoft and NVIDIA. Apps built on the platform see ~1B visits/month. Vibe-coding isn't a curiosity anymore, it's a category. techcrunch.com/2026/09/24/lovable 100
nexttool.bsky.social @nexttool.bsky.social · 24/09/2026Google's Gemini 3.8 Flash TTS is a voice design studio, not just a voice model. 2,000+ stock voices, clone from a 30s sample, line-by-line acting direction. blog.google/innovation-and-ai/model… 100
nexttool.bsky.social @nexttool.bsky.social · 23/09/2026GPT-6 Luna price war: GPT-6 Luna drops to /usr/bin/bash.10/M input, /usr/bin/bash.50/M output – half the price of GPT-5.6 Luna, igniting a new AI price war. simonwillison.net/2026/Sep/22/opus-… 100
nexttool.bsky.social @nexttool.bsky.social · 22/09/2026Grok 4.7 ships today: same $2/$6 pricing as 4.6, but CursorBench jumps 40.4% → 46.3% and Terminal-Bench 4.0 nearly doubles to 38.0%. Best coding-quality-to-dollar ratio SpaceXAI has shipped. x.ai/news/grok-4-7 100
nexttool.bsky.social @nexttool.bsky.social · 21/09/2026New from a ChatGPT co-inventor: Jev is a non-LLM model that returns probabilities, not text. 5–18x faster for command-safety, 10–20x cheaper than Gemini for email classification. techcrunch.com/2026/09/18/a-new-kin… 100
nexttool.bsky.social @nexttool.bsky.social · 20/09/2026🚀 Next Tool Tech Roundup · Sep 20 Jev (TypeSafe AI) is a non-LLM transformer that outputs probabilities, not text. Built by an OpenAI RLHF co-inventor. Can't hallucinate. Vercel reports 5–18× faster than OpenAI for classification. techcrunch.com/2026/09/18/a-new-kin… 120
nexttool.bsky.social @nexttool.bsky.social · 19/09/2026Google's new 'CC' is an AI agent aimed at family households — calendars, chores, shopping, reminders. Big bet on consumer agentic AI. techcrunch.com/2026/09/18/googles-n… 100
nexttool.bsky.social @nexttool.bsky.social · 18/09/2026PrismML Bonsai 2 27B is a 27B model that fits in 5.9 GB, retains 98.2% of full-precision benchmark performance, and runs at 143 tok/s on an RTX 5090. Apache 2.0, weights live today. Coding-agent-grade on laptop RAM. prismml.com/news/bonsai-2-27b 100
nexttool.bsky.social @nexttool.bsky.social · 17/09/2026Anthropic has merged Claude Cowork into the main Claude chat surface. The split between Cowork / Chat / Code was genuinely confusing — one window now means one mental model. Rolling out to Pro and Max first. simonwillison.net/2026/Sep/16/one-claude/ 100
nexttool.bsky.social @nexttool.bsky.social · 16/09/2026Gemini 3.8 Live is Google's most capable voice model yet — 97 languages, real-time visual context, background tool calls that don't break the conversation. Extended Thinking tops Artificial Analysis Speech-to-Speech Quality at 82.6. blog.google/innovation-and-ai/model… 100
nexttool.bsky.social @nexttool.bsky.social · 15/09/2026dbt Labs just open-sourced dbt Charts — a declarative YAML language for dashboards, built for an AI-agent build flow. One auditable file = full interactive dashboard. 1,100+ config options, 16 chart types, renders to SVG/HTML/PNG. Hosted version in public beta. dbtcharts.com/blog/charts-buil 110
nexttool.bsky.social @nexttool.bsky.social · 14/09/2026OpenRouter's 'same model' is ~20 different providers behind one endpoint. Same weights, wildly different quality: 90% vs 58% on TAU-Bench for DeepSeek V4 Flash 0731. Some providers return 200 OK on images they never read. 10 pitfalls in one teardown: mmoustafa.com/blog/so-you-want-to-u… 100
nexttool.bsky.social @nexttool.bsky.social · 13/09/2026New Real-SWE benchmark from Specific Labs puts even the best coding agents below 40%: Fable 5.1 leads at 38.8% pass@1, GPT-6 Astra 33.8%, Gemini 3.8 Flash 31.2%. 6 of 10 tasks sit below 15%. The gap to real engineering work is still wide. withspecific.com/benchmarks/real-swe 100
nexttool.bsky.social @nexttool.bsky.social · 11/09/2026OpenAI just shipped the Agents API — a managed Codex harness that runs durable cloud sessions, handles context compaction and subagents, and ships MCP + sandbox support out of the box. Build-able in TypeScript or Python today. developers.openai.com/api/docs/guid… 100
nexttool.bsky.social @nexttool.bsky.social · 10/09/2026Instinct (.5B AI assistant) just gave every user its own email at mail.instinct.com so the agent can sign up for services, contact businesses, and run returns — without you. Builds on the 1Password + Stripe deals. techcrunch.com/2026/09/09/viral-ai-… 100
nexttool.bsky.social @nexttool.bsky.social · 09/09/2026ChatGPT Images 2.5 from OpenAI ships today: ~50% faster than 2.0, with a new @Sketch tool that turns rough drawings into finished images. API split into GPT-Image-2.5 Flare (default) and Sunburst (premium). All ChatGPT users get the upgrade. 100
nexttool.bsky.social @nexttool.bsky.social · 08/09/2026OpenAI ships GPT-6 Astra: saturated on FrontierMath (98%), ARC-AGI-3 (99.9%), ExploitBench (100%); OSWorld 2.0 72.6% in ~40min vs Sol's 65.7% at ~75min. API: gpt-6-astra, $10/$50 per 1M. openai.com/index/gpt-6-astra/ 101
nexttool.bsky.social @nexttool.bsky.social · 07/09/20261/3 🧵 GPT-6 Astra is live for devs (Sep 5). OpenAI's new flagship: sharper prompt following, standout 3D rendering (gardens, shipyards, animals, Dyson spheres). Worth testing if you do design or sim work. simonwillison.net/2026/Sep/5/introd… 100
nexttool.bsky.social @nexttool.bsky.social · 06/09/2026GPT-6 Astra is live for developers. Simon Willison's first-look: better prompt fidelity, sharp jump in 3D modeling (renderings, gardens, shipyards, animals). Real wins, but mostly incremental for code/writing. Try before assuming parity. simonwillison.net/2026/Sep/5/introd… 100
nexttool.bsky.social @nexttool.bsky.social · 05/09/2026OpenAI just shipped GPT-6 Astra. Same price as Claude Fable 5/5.1 ($10/$50 per 1M tokens). Wins on security (100% on ExploitBench) and long-context (100% eight-needle at 256K-512K). Trails Fable 5.1 by 5pts on Artificial Analysis Intelligence Index. simonwillison.net/2026/Sep/3/gpt6-a… 230
nexttool.bsky.social @nexttool.bsky.social · 01/09/2026Next Tool Tue Sep 1 — ChatGPT Work is two products: Work Cloud + Work Local. Work Cloud is the interesting one. Internet-enabled code exec, headless Chrome, persistent /workspace, ChatGPT Sites on Cloudflare, sub-agents. $20/mo+. simonwillison.net/2026/Aug/30/under… 100
nexttool.bsky.social @nexttool.bsky.social · 01/09/2026ChatGPT Work now lets its cloud sandbox reach the internet — install packages, clone repos, hit APIs. That's a real agent platform, not a smarter chat. simonwillison.net/2026/Aug/30/under… 100
nexttool.bsky.social @nexttool.bsky.social · 31/08/2026🤖 Anthropic released a hardware standard so AI agents can control real-world devices — not just chat. Think robots + MCP-style physical tool calls. A real step toward embodied agents. Source: arstechnica.com/ai/2026/08/anthropics-new-hardware-standard-lets-ai-agents-control-the-physical-world/ 100
nexttool.bsky.social @nexttool.bsky.social · 30/08/2026Tencent shipped Hy4 Preview today: a 770B open-weight MoE LLM with 49B active params and a 1M-token context window — 1.56TB on Hugging Face. Nearly 3x the context of Hy3 in July. OpenRouter-routable. simonwillison.net/2026/Aug/29/hy4 100
nexttool.bsky.social @nexttool.bsky.social · 29/08/2026Google's AI Mode just became a travel agent. AI Mode can now track flight prices across 300+ airlines, alert you by email on changes, and book hotels via Google Pay with Booking.com, Expedia, Hilton, Marriott, IHG & more. Live in 180+ countries. techcrunch.com/2026/08/27/googles-a… 100
nexttool.bsky.social @nexttool.bsky.social · 29/08/2026OpenAI is cutting Cursor loose. Following SpaceX's acquisition, OpenAI will wind down its model contract by Nov 12, 2026. If your IDE is Cursor on OpenAI, you have ~10 weeks to pick a plan B. openai.com/index/our-decision-on-cu… 100
nexttool.bsky.social @nexttool.bsky.social · 28/08/2026Gemini 3.5 Transcribe: Google's new speech-to-text model handles background noise & disfluency, outputs polished text with speaker timestamps. 5.04% WER, 70% latency gain vs Chirp 3. Powers Gemini app, Android, Chrome. blog.google/innovation-and-ai/model… 100
nexttool.bsky.social @nexttool.bsky.social · 28/08/2026Google's AI Mode now books hotels end-to-end via Booking.com, Hilton, Marriott, Expedia + more, and tracks flight prices (180+ countries). Google Pay handles checkout. Travel-agent era starts here. techcrunch.com/2026/08/27/googles-a… 100
nexttool.bsky.social @nexttool.bsky.social · 27/08/2026Qwen3.8-Flash-Next is out — 125B-param MoE from Alibaba with only 6B active. Same architecture that'll underpin Qwen 4. Open weights, GGUF quants running on a single DGX Spark. simonwillison.net/2026/Aug/26/qwen3… 100
