16 comments

  • redmalang 6 hours ago
    We have an internal proxy (that I've been meaning to open source for ages) that routes all llm usage at our company, which allows us to see data in realtime. Its been fascinating how rapidly Pi has been adopted. Moreover since its pretty hackable, we've been able to automatically aggregate context from pi sessions, which has resulted in Pi efficacy being higher as more people use it, putting in place a interesting virtuous loop. I didn't expect this outcome: for whatever reason I assumed proprietary harnesses fine tuned to work with a companies' models would work better? ps/random aside: there is something slightly off about Pi's edit command, we are planning to investigate this further and patch this as we have quite a few session traces now..
    • Schlagbohrer 16 minutes ago
      Can you explain context aggregation here more? What is that? What are you doing and how is it helping?
    • lukax 5 hours ago
      Yes, this is a known issue. A significant amount of Edit tool calls fails in Pi witg newer models.

      https://lucumr.pocoo.org/2026/7/4/better-models-worse-tools/

    • redmalang 1 hour ago
      Aside 2: Anecdotally we found that Pi performs more or less on par with native harnesses at lower cost on decently specified prompts. It is also phenomenal at context cacheing especially on Deepseek models (its hard to precisely attribute credit here are my understanding is this is a DS speciality). But it fails much worse on poorly drafted prompts. I'm generalising but native harnesses seem to be better kind of flailing along on those.
      • swingboy 2 minutes ago
        Is it because Pi’s default system prompt is so simple?
    • Sphax 4 hours ago
      > we've been able to automatically aggregate context from pi sessions

      what do you mean by this ? do you rewrite the context in your proxy ?

      • N_Lens 4 hours ago
        I imagine they're capturing all the tokens+context from pi sessions from every developer.
      • redmalang 4 hours ago
        Yeah.
        • pbowyer 44 minutes ago
          Can you say more? This sounds like it could be more useful than trying to build a knowledge base (say qmd) and then get the models to search it for the right context each time.
          • redmalang 5 minutes ago
            This is in fact what we do (with higher order abstractions now built on top of this). This builds self evolving interactive knowledge base and puts it into a QMD searchable index. The indexer is already open source: https://github.com/jibs/duffel
  • zipy124 10 minutes ago
    I'd like to know what their config with pi is like. Is it vanilla, is it oh my pi etc.... This seems like important info.
  • anentropic 2 hours ago
    > the results showed clear clustering of the models and harnesses into 3 capability tiers

    pretty sure the only thing making that 'clear' is the coloured stripes, if you took that away it'd look like two tiers

    good result for GLM 5.2 though

    and Sonnet 5 seems like a waste of time

  • Schlagbohrer 17 minutes ago
    Anthropic is not beating the charges that they inflate token consumption with their own harness given these findings that Pi is 2.2x more efficient at token management. Big "toothpaste ads tell you to use way too much toothpaste" energy.
  • yodon 7 hours ago
    I wish they'd do a follow-on post drilling into the impact of the programming language on cost-per-task, specifically looking at cost to complete tasks in mainstream strongly typed languages (eg. C#, TypeScript) vs dynamic languages (eg. Python, JavaScript). Does the additional verbosity of the language help or hurt cost per task?
    • trollbridge 4 hours ago
      I don't have hard data, but we have shifted to Rust and Swift (for frontend UI) for the bulk of our dev simply because it is a lot more predictable, easier for tool calls to edit, the build steps produce easier output for the agent to loop on, the tests are easier to write/get results from, etc., although I am mostly measuring this in time, not cost.

      Once the thing is rock-solid it's relatively easy to do a Swift->HTTP/HTML/CSS/React/TypeScript conversion.

    • rootatixww3 2 hours ago
      for that you would need to compare the same task implemented in two different languages - C# and Python for example, no?
  • lukax 5 hours ago
    Could it be that users of Pi are more senior and know better how to prompt and that's why the pass rate is higher?
    • rootatixww3 2 hours ago
      they explain this is a benchmark, all models/harnesses receive the same prompt
  • cpard 7 hours ago
    This was mostly because Sonnet 5 worked longer and read more to get there, consuming 1.9x more tokens.

    I have experienced similar behavior between opus and haiku when benchmarking Dara engineering tasks. The “cheaper” model takes many more turns to figure out the task and this is without taking into account other important factors.

    Another interesting behavior that I observed is that Haiku tended to cheat more maybe because it was having a harder time to find the root cause of the problem.

    Benchmarking and evaluation of agentic systems is very interesting and if there’s one thing that someone should keep from the Databricks post is how important is for everyone to build and run their own.

  • yigitcan07 2 hours ago
    Would be great to see time spent per task per model. Especially since article references 390+ tokens per second for GLM5.2.
  • falaki 10 hours ago
    1) Many models are now competitive at the top tier, including open source. 2) GLM 5.2 in particular was a major step forward in open source coding agent performance, 3) Harnesses make a huge difference in cost-performance. 4) Cheaper per-token does not imply cheaper per-task.
    • falaki 10 hours ago
      Also they suggest every company should build their own benchmark and repeat these tests with new models instead of relying on the SWE bench.
      • fmind-dev 3 hours ago
        It takes time and effort to build such benchmark. It works at Databricks scale, I'm not sure smaller companies are ready to invest on internal benchmarks.

        But they are more vendor neutral, now they don't sell their own model. It's interesting from a benchmark point of view.

  • pianopatrick 2 hours ago
    Seems like for a hobby project $1 or $2 per task would add up a bit, depending on how many tasks you need to do. I mean it makes sense for a software company
  • jkwang 3 hours ago
    The repo-scale angle is the useful part here. Small synthetic tasks miss a lot of the integration and context retrieval failures you only see in a codebase this large.
  • dirkc 1 hour ago
    > as we aggressively adopt AI for engineering

    Why do we need to aggressively adopt things rather than thoughtfully adopt things?

    It sounds like they are probably punching AI and engineers in the process

    • simianwords 1 hour ago
      What if other competing companies adopt it aggressively and push out features faster and get more market share?

      What if pushing features faster brings more money because users like the features?

  • throwa356262 2 hours ago
    Is there any technical analysis of why contex grows slower in Pi compared to codex and CC?
    • rootatixww3 2 hours ago
      different system prompts, codex will have system prompt telling the model to gather a lot more context before starting work
  • zkmon 6 hours ago
    > Databricks’ multi-million line codebase

    The combined size of codebases for the underlying opensource products (Apache Spark etc) might be around 1M lines, I think. Why does the orchestration/management layer, that is "databricks", exceed the sizes of the core products?

    • trollbridge 4 hours ago
      Because if they're like previous places I worked out with proprietary commercial codebases, the code is old, has been around a long time, and trying to shrink it is not a priority.

      It's a good stress test for the LLM because it is not an "ideal" codebase.

    • alexott 3 hours ago
      :facepalm: "orchestration/management layer" that just runs dozens of millions of VMs per day, across the three clouds, with things like model serving, AI gateway, OLTP (Lakebase/Neon), Unity Catalog, etc.

      Forget Databricks == Apache Spark...

    • ozgrakkurt 5 hours ago
      Old codebase, you always add code and never remove it. So it is expected to be like this.

      Deleting code is difficult and almost never makes sense afaik

      • tijs 4 hours ago
        Every line you delete is a line you no longer need to maintain. We aggressively prune old code in our apps and it has definitely helped with maintainability. For a mobile app it’s also code you don’t ship so that’s a nice bonus which I guess is not much of an argument on a backend codebase
      • wwind123 4 hours ago
        On one hand, I understand that some old code is hard to delete because it's hard to detangle a lot of the legacy dependency. On the other hand, too much useless old code existing in the code base by itself could become a big maintenance burden for both humans and AI. In some cases at some point it might become more economical to just invest a bunch of resource to detangle the dependencies to be able to remove the old code.
        • hasyimibhar 1 hour ago
          At huge companies, it's hard to prioritize this because it's hard to pin dollar values to removing legacy code, while it's easier to show how building feature X will earn the company $Y amount of revenue. And because of that, there is also no incentive to do it, you don't get promoted by deleting old code, you get promoted by showing how your effort helped contribute to company revenue. At my previous company (100+ engineers, hundreds of microservices), teams that regularly clean legacy codebase tend to be platform teams (cost centers), while teams that struggle to get these prioritized are product vertical teams (revenue centers).
    • appplication 5 hours ago
      LoC isn’t a super helpful metric so I think the better question is why is the headline using it. I can say I’ve personally created about 200k LoC code in the last 5 years and most of that has some value. But it really doesn’t say might about how much value or really anything else meaningful.
      • InsideOutSanta 4 hours ago
        They're probably mentioning the size of the code base as an indicator for how difficult it is for an LLM to understand where and how to make changes.
  • vegetablefinger 7 hours ago
    [flagged]