Top all-time · #3780
Quesma
Company blog · quesma.com/blog · on HN as staredaiprogramming
Quesma is the company blog at quesma.com, writing about ai and programming. Since 2023, 20 different Hacker News readers have linked it 130 times, most often “Nano Banana Pro: raw intelligence with tool use”. It is still writing: the last post was on 2026-10-07, and you can follow it by feed.
Too few sampled comments (9) to say how often HN recommends it.?
Why HN links it
Commenters link Quesma posts and benchmarks on AI models, LLM token costs, binary analysis with AI, and data tools like ClickHouse and Iceberg. The links are cited as sources or examples, without judging the writing.
Summarised by Claude from a sample of 9 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Connect Claude Code, Codex and Cursor session costs to engineering work and outcomes. Find wasted tokens, repeated failures and practices worth sharing.
Most-linked posts?
- Nano Banana Pro: raw intelligence with tool use3 people · 5 mentions · first linked 2025-11-24
- I gave Opus 5.5 one prompt and six hours to visualize Invisible Cities3 people · 3 mentions · first linked 2026-10-07
- BinaryAudit: Can AI find backdoors in raw machine code?2 people · 8 mentions · first linked 2026-02-11
- Baba Is Solved by Fable 5 and GPT-5.6 Sol, but at what cost?2 people · 6 mentions · first linked 2026-07-16
- We hid backdoors in binaries – Opus 4.6 found 49% of them2 people · 6 mentions · first linked 2026-02-10
- AI for coding is still playing Go, not StarCraft2 people · 4 mentions · first linked 2025-10-14
- Don't Let Apache Iceberg Sink Your Analytics: Practical Limitations in 20252 people · 3 mentions · first linked 2025-05-22
- Agent session transcripts are precious, keep them2 people · 2 mentions · first linked 2026-10-01
- The true cost of saying "Hi" to an AI agent2 people · 2 mentions · first linked 2026-07-09
- Which chart would you swipe right2 people · 2 mentions · first linked 2025-09-08
What Hacker News says?
Qwen3.8 27B (which I adore) is nowhere near Opus 4.8 at puzzle games testing fluid intelligence, https://quesma.com/blog/baba-is-aug-2026/
Comment linking /blog/baba-is-aug-2026/
on “GLM-5.3-Flash”Submitted I gave Opus 5.5 one prompt and six hours to visualize Invisible Cities as a story · 418 points
HN discussionIf you want to combine Claude Code coding with reasoning, it is easy to do it with a plugin - https://github.com/stared/gemini-claude-skills , wrote for myself, but shared in case anyone wants. Somehow bigger context here: https://quesma.com/blog/claude-skills-not-antigravity/ .
Comment linking /blog/claude-skills-not-antigravity/
on “GPT-5.2-Codex”Submitted Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses as a story · 286 points
HN discussionFLUX.1 Pro Kontext was one of the best artistic model, still great at instruction following comparing to MidJourney V7. See my third comparison in Nano Banana blog post: https://quesma.com/blog/nano-banana-pro-intelligence-with-to...
Comment linking /blog/nano-banana-pro-intelligence-with-tools/
on “FLUX.2: Frontier Visual Intelligence”Submitted Tau² benchmark: How a prompt rewrite boosted GPT-5-mini by 22% as a story · 197 points
HN discussionKeep in mind these downloadable models use 3-10x the amount of tokens as well. You really can’t beat a couple $20 subscriptions. https://quesma.com/benchmarks/babaisbench/
Comment linking /benchmarks/babaisbench/
on “My local model setup on an M4 Pro Mac Mini”Submitted RTK reports token savings, but our cost benchmarks disagree as a story · 170 points
HN discussionYes! I was raving about Claude Skills a few days ago (vide https://quesma.com/blog/claude-skills-not-antigravity/ ), and excited they come to Codex as well!
Comment linking /blog/claude-skills-not-antigravity/
on “Skills Officially Comes to Codex”Submitted The security paradox of local LLMs as a story · 160 points
HN discussionIceberg competitor, addressing some of its shortcomings, like blown-up metadata: https://quesma.com/blog-detail/apache-iceberg-practical-limi... Even Snowflake was using FoundationDB for metadata, whereas Iceberg attempts to use blob storage even for the metadata layer.
Comment linking /blog-detail/apache-iceberg-practical-limitations-2025
on “DuckLake is an integrated data lake and catalog format”Submitted CompileBench: Can AI Compile 22-year-old Code? as a story · 148 points
HN discussionThank you for the feedback! In this (telecom) benchmark you can review agent policies and manuals here: 1) https://github.com/sierra-research/tau2-bench/blob/main/data... 2) https://github.com/sierra-research/tau2-bench/blob/main/data... Of course these are just parts of the prompt, you can inspect benchamark code to see how these are rendered to actual LLM calls. In case someone is not familiar with framework metho…
Comment linking /blog/tau2-from-llm-benchmark-to-blueprint-for-testing-ai-agents/
on “Tau² benchmark: How a prompt rewrite boosted GPT-5-mini by 22%”Submitted Qwen 3.6 27B is the sweet spot for local development as a story · 117 points
HN discussionLLMs are wonderful for migration. Also, are good at exploring APIs. A month ago I migrated company's website and blog from Framer to Astro ( https://quesma.com/blog/ is you would like to see the end result). This weekend I created a summary of Grafana dashboard data. LLMs are tireless at checking hypothesis, running grunt code, seeing results, and iterating on that. What takes more than a single is to check if the r…
Comment linking /blog/
on “Notion API importer, with Databases to Bases conversion bounty”Submitted Mushroom hunting with LLMs: what can go wrong? as a story · 56 points
HN discussionWhile I like this index, calling in "Intelligence" might be confusing - it is a mix of coding and knowledge. Compare and contrast with ARC-AGI, BabaIsBench ( https://quesma.com/benchmarks/babaisbench/ ), or MazeBench ( https://mazebench.com/blog?post=introducing-mazebench ). In particular, in one Baba Is Bench post ( https://quesma.com/blog/baba-is-aug-2026/ ), while quoting a Pareto frontier chart from AA, I noted:…
Comment linking /blog/baba-is-aug-2026/
on “Artificial Analysis Intelligence Index v4.2”Submitted Tau² Benchmark in Action: Early Results and Key Takeaways as a story · 16 points
HN discussionIf you want to use it to create ggplot2 charts, here is an open source demo: https://github.com/QuesmaOrg/demo-webr-ggplot/ I created it as a side project from using WebR to execute code LLM-generated code ( https://quesma.com/blog-detail/sandboxing-ai-generated-code-... ). While we migrated away from it, I saw that WebR is cool, and I wanted to share it with you.
Comment linking /blog-detail/sandboxing-ai-generated-code-why-we-moved-from-webr-to-aws-lambda
on “WebR – R in the Browser”Submitted Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses as a story · 15 points
HN discussionKibana is great, but managing Elasticsearch is pain. Slow ingestion, query performance, takes a lot of space, harder to tune up. Modern columnar SQL such as ClickHouse are 10+ times more efficient in real-world use cases. I'm a CEO and founder of Quesma, which, let's use Kibana with ClickHouse: https://quesma.com/ Forever free, source-available license.
Comment linking /
on “Some notes on Grafana Loki's new "structured metadata"”Submitted I burned all my tokens researching how to save tokens as a story · 11 points
HN discussionIs it possible to re-run it? I am curious for Gemini 3 Pro. As a side note, it is easy to create sharable experiments with Harbor - we migrated our own benchmarks there, here is our experience: https://quesma.com/blog/compilebench-in-harbor/ .
Comment linking /blog/compilebench-in-harbor/
on “AI Coding Assistants Are Getting Worse”Submitted Outside of the bubble, AI is Black Mirror as a story · 7 points
HN discussionMoreover, models are becoming more aggressive in their use of tools for simple tasks. E.g., saying "Hey" or "commit" can trigger 30+ tool uses: https://quesma.com/blog/the-true-cost-of-saying-hi-to-an-ai-... Tokenflation seems very real. The number of tokens needed for simple things is increasing.
Comment linking /blog/the-true-cost-of-saying-hi-to-an-ai-agent/
on “Claude Code sends 33k tokens before reading the prompt; OpenCode sends 7k”Submitted Migrating CompileBench to Harbor: standardizing AI agent evals as a story · 6 points
HN discussionFor malware detection, many models are biased for or against detecting a threat (likely a thing that can be adjusted with a prompt). I suggest tasks cannot be guessed (find, not tell). And 2d charts, both for ROC and pricing, vide https://quesma.com/benchmarks/binaryaudit/
Comment linking /benchmarks/binaryaudit/
on “Will It Mythos?”Submitted A postmortem on our $2.5M database gateway: lessons from pilot purgatory as a story · 6 points
HN discussionIn https://quesma.com/blog/nano-banana-pro-intelligence-with-to... (Nov 2025) we had an illustrative diagram of using Nano Banana Pro to create a circuit diagram.
Comment linking /blog/nano-banana-pro-intelligence-with-tools/
on “My smart sleep mask broadcasts users' brainwaves to an open MQTT broker”Submitted Don't Let Apache Iceberg Sink Your Analytics: Practical Limitations in 2025 as a story · 6 points
HN discussion
