TiTikey

GPT 5.5 is much smarter than I thought Yesterday, I did one-shots, coding, benchmarks, and was disappointed. Today, I did it all again, except via the API, which is now available. Results changed completely: → one-shot prompts went from bad to very good → excellent coding outputs, on both pi and holefill → benchmarks jumped, and now GPT *dominates* I don't know what happened, I suppose there is something wrong with my Codex. In any case, truth is this model is very smart. It obliterated my benchmark, which is crazy because some of these problems were meant not to be solved. I'll need much harder tasks. I also fixed 2 bugs that affected some providers: → added a retry for lost connection → removed the timeout limit DeepSeek and Kimi wanted to spend more than 1 hour on my prompts, so I let them. Their results are much better now. Kimi K2.6 almost reaches Sonnet 4.6, although much slower. Also this shows my points from last post were wrong Again: this is a new vibe-coded bench, I'm focused on other things, so expect bugs and don't over-read this! GLM 5.1, Gemma, Grok are not updated yet.

We benchmarked GPT-5.5 on document understanding 📄📊 We ran it through ParseBench, our comprehensive OCR benchmark over enterprise documents. We evaluated metrics across various dimensions: visual grounding, tables, charts, and more. We evaluated GPT-5.5 on mid thinking and zero-thinking modes. When compared against GPT-5.4 (0 thinking) and Opus 4.7 (adaptive thinking): 📈 GPT-5.5 wins on tables 📈 GPT-5.5 wins on visual grounding 📉 GPT-5.5 0-thinking does worse on charts than GPT-5.4 0-thinking 📉 Higher thinking does worse than lower thinking of content faithfulness, semantic formatting 📉 Opus 4.7 wins overall on content faithfulness and semantic formatting 💸 GPT-5.5 is expensive: 13c per page at mid-thinking modes and 5.93 at 0-thinking! This is 5x the cost of any competitive OCR solution. Conclusion: GPT-5.5 is one of the better frontier models out there in terms of pure accuracy, but def not pound for pound w.r.t price.

We benchmarked GPT-5.5 on document understanding 📄📊 We ran it through ParseBench, our comprehensive OCR benchmark over enterprise documents. We evaluated metrics across various dimensions: visual grounding, tables, charts, and more. We evaluated GPT-5.5 on mid thinking and zero-thinking modes. When compared against GPT-5.4 (0 thinking) and Opus 4.7 (adaptive thinking): 📈 GPT-5.5 wins on tables 📈 GPT-5.5 wins on visual grounding 📉 GPT-5.5 0-thinking does worse on charts than GPT-5.4 0-thinking 📉 Higher thinking does worse than lower thinking of content faithfulness, semantic formatting 📉 Opus 4.7 wins overall on content faithfulness and semantic formatting 💸 GPT-5.5 is expensive: 13c per page at mid-thinking modes and 5.93 at 0-thinking! This is 5x the cost of any competitive OCR solution. Conclusion: GPT-5.5 is one of the better frontier models out there in terms of pure accuracy, but def not pound for pound w.r.t price.

You NEED to run this morning briefing with Claude Cowork every weekday: 1. Reads your 3 context files first (about-me, brand-voice, working-style). 2. Pulls today's Google Calendar events with times, attendees, and prep notes flagged. 3. Scans Gmail from the last 24 hours. Flags anything from clients or partners. Flags anything that needs a reply. 4. Organizes everything by urgency. Action Today vs. FYI Only. 5. Saves the full briefing to your project folder. Review anytime. Set it once. It runs on autopilot every weekday. One catch: scheduled tasks only run when your computer is on and the desktop app is open. Run it manually the first time after setup. That pre-approves Gmail and Calendar so future runs never pause waiting on permissions.

GPT-5.5 and GPT-5.5 Pro are now available in Hermes Agent through the Nous Portal and OpenRouter providers! (alongside the direct openai oauth provider from yesterday)

Marketing Skills v1.9.0 is out. What shipped: 🆕 /image — AI image generation for marketing. Gemini, Flux, Ideogram, DALL-E, Midjourney. Blog heroes, social graphics, product mockups, and optimization workflows. 🆕 /video — AI video production. Hyperframes & Remotion pipelines, HeyGen & Synthesia avatars, Veo, Runway, Kling, Pika generation. Editing, repurposing, and distribution. Enhanced skills: • /social-content — short-form video section (TikTok, Reels, Shorts frameworks with hooks and scripting) • HeyGen integration — API setup, MCP server, avatar workflows • Hyperframes integration — HTML/CSS programmatic video rendering 📹 This launch video was made with Hyperframes! Also: plugin marketplace fix, phishing URL removal. Now shipping 40 skills and 52 tool integrations. Free, open source. npx skills add coreyhaines31/marketingskills

X Square Robot unveiled WALL-B, a next-generation embodied AI foundation model. The company announced an aggressive timeline, stating these general-purpose robots will begin deploying into real-world households within exactly 35 days.

I TESTED GPT 5.5 AGAINST OPUS 4.7 AND NEED YOUR HELP IN DETERMINING A WINNER Landing Page Winner: Codex, much prettier, much better Creating AI UGC with Fal AI PI: Opus 4.7, much more realistic and cleaner iOS App Winner: IDK HELP ME ↓

The 5-minute security audit most Claude Cowork users skip: 1. Data sensitivity: pick "mostly business content" (docs, email, calendar). Not PII, not financial data. 2. External actions: pick "ask before any external action." Not just sensitive ones. All of them. 3. Scheduled tasks: pick "read-only tasks auto-run." Anything that writes still needs your approval. 4. External docs: pick "rarely, mostly my own content." Claude adjusts its trust level accordingly. Cowork then scans your workspace to confirm zero sensitive patterns before writing the rules. Do this when done: add the generated rules to your global instructions so they load every session. That one step turns your 4 answers into a permanent policy. No more re-approving the same actions every time you open Cowork.

GRPO, explained visually: (learn LLM fine-tuning with GRPO below)

Sam Altman: “If I were 22 right now, I'd feel like the luckiest kid in history." 60 minutes dissecting the brain of the OpenAI founder. An interesting watch.