Open TTS Leaderboard: Scalable Evaluation for Multilingual Text-to-Speech and Voice Cloning

GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price

• <p><a href="https://news.ycombinator.com/item?id=49896586#49898129">My comment</a> on <a href="https://news.ycombinator.com/item?id=49896586">GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price</a> &mdash; Hacker News.</p><p>I'm a bit late with the pelicans because I was live-blogging the

OpenAI slashes GPT-6.1 Sol costs by 80% while new Decisions API routes app logic in 150ms, signaling a broader industry shift toward cheap, fast infrastructure over raw model power.

• OpenAI's GPT-6.1 Sol matches flagship Astra on coding/compute tasks at ~1/5 to 1/7 the cost, with cached input now $0.10/M tokens (half the previous price)
• OpenAI's new Decisions API uses a tuned GPT-6 Luna model to classify/route app logic in 150ms vs 1.6s for standard calls — built for content moderation, ticket routing, and agent decision-making
• Meta published a training trick that doubles Qwen3-8B's math accuracy while shortening output length, free to apply
• PageIndex (open-source) hits 98.7% accuracy on FinanceBench using a tree-search document structure instead of vector embeddings/chunking — no vector DB required
• Anthropic's Claude Sonnet 5.5 matches Opus 5.5 on agentic benchmarks but burns 7x more tokens to get there
• Blackfrost AI released a 180B MoE model with only 6B active params in GGUF format, continuing the trend toward cheaper inference-efficient architectures

OpenAI's DevDay flood of agent tools and pricing shifts matters more for builders than the White House's toothless AI safety pledge.

• OpenAI launched 'Dots' at DevDay: always-on cloud agents powered by GPT-6 Astra, working across Slack/Teams/Messages, available only to Pro ($500/mo tier) and Business Premium users.
• New GPT-6.1 Sol model offers 'near Astra intelligence' at a fifth of the price ($2/M input, $10/M output tokens), aimed at agentic coding and professional work.
• Pro plan ($200/mo) usage limits were cut: Work/Codex allowance dropped from 20x to 10x Plus tier, and GPT-6 Pro weekly messages halved from 200 to 100.
• New Ultrafast tier gives 8x speed in Codex and 6x in API; OpenAI also launched a Marketplace (32 partners incl. Adobe, Figma, Harvey) letting enterprise token spend buy third-party software.
• Six AI CEOs (Amodei, Pichai, Zuckerberg, Brockman, Musk, Huang) signed a White House 'Accord on Superintelligence' pledging self-regulation—no new laws, just internal audits and honor-system oversight.
• Anthropic warned GLM-5 can autonomously build cyber exploits, while Meta rolled out Muse for Small Business, intensifying the consumer-agent race OpenAI is explicitly avoiding.

Google announces Gemini 4 Argon AI model, but you can't use it yet

AMD acquires World Labs AI startup, upping the ante against Nvidia

OpenAI DevDay 2026 live blog

• <p>I'm at <a href="https://devday.openai.com/">OpenAI DevDay</a> today, in Fort Mason, San Francisco. Same as <a href="https://simonwillison.net/2025/Oct/6/openai-devday-live-blog/">last year</a> I'll be live blogging the keynote and some other notes during the day.</p> <p><em>OpenAI gave me a free

Claude Sonnet 5.5 quietly leapfrogs Opus 5.5 on coding benchmarks while cutting costs 30%, and OpenAI fixes a silent vision bug that was degrading GPT-6 image results all along.

• Anthropic's Claude Sonnet 5.5 scores 70.6% on Terminal-Bench 4.0 (vs 10.3% for Sonnet 5, 66.4% for Opus 5.5), same price but 30% faster/cheaper, with 1M token context and 128K output
• OpenAI silently patched a bug in GPT-6 Sol and GPT-6 Luna that was degrading image understanding across API, Codex, and computer-use tasks — worth rerunning any vision pipeline tests now
• xAI shipped Grok Team Bots: shared, persistent bots in Slack that teams configure once and everyone uses, with pre-built templates for sales/marketing/product/data
• CodeAF, an open-source coding harness (Apache 2.0), claims 4x more GitHub issue solves than Claude Code on the same open model (DeepSeek/Qwen/GLM/Kimi) at half the cost
• Google's autonomous research agent reportedly outperformed human-written papers by 25% across 86 tasks, and AMD acquired World Labs, making Fei-Fei Li its chief scientist

Anthropic undercuts OpenAI on price with Claude Sonnet 5.5 while Meta's enterprise AI push collides with mounting Muse privacy scandals

• Anthropic launched Claude Sonnet 5.5: 30% faster, 30% cheaper on tokens at same $2/$10 per million price as Sonnet 5, beats Sonnet 5 on all benchmarks and even outscores Opus 5.5 on agentic coding (Terminal-Bench 4.0)
• Neolabs like Jev (Typesafe) and Pathway are emerging as threats to frontier labs by promising comparable performance at much lower cost using non-LLM architectures
• Meta launched Meta Enterprise Platform bundling Muse, Business Agent, Muse API and Muse Code, hiring MongoDB's CJ Desai as Chief Enterprise Platform Officer, right as Muse faces serious privacy incidents (leaked home address to a stranger, unauthorized iMessage access, a scrapped 'human concierge' feature where a contractor made a racist remark on a user's behalf)
• Nvidia unveiled Open Agent Safety (OpenShell + Sentry) to contain rogue agents in milliseconds, with 100+ partners including Anthropic, Microsoft, SpaceX - but OpenAI, Google, Meta and Amazon are absent
• Anthropic's IPO prospectus revealed a $42 billion net loss for 2025, underscoring how expensive the frontier race remains even as per-token prices drop

OpenAI says planned GPT-6.1 is too insecure to release

Quoting Anthropic Frontier Red Team

• <blockquote cite="https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities"><p>We evaluate several models on 100 tasks from the [internal Binary Exploitation benchmark] (selected at random), and find that GLM-5.3 develops full control flow hijacks in 4% of the trials;