gpt-6-luna real-task evaluations

Every report keeps its original task, method, finding, limitation, evaluator identity and verification date. A community report is evidence about that run, not a universal score.

  • GPT-6 Astra, Sol and Luna on one SVG animation prompt

    By 雪瑜 (@xueyu1125). Programmer and AI-tools builder on X who posted the timings and quota readings Published 2026-09-23; retrieved 2026-09-24; verified 2026-09-24.

    Task
    Generate an SVG animation of a pelican riding a bicycle, shown in H5
    Method
    Same prompt on GPT-6 Astra, GPT-6 Sol, GPT-6 Luna and GPT-5.6 Sol; the author recorded time taken and the share of a five-hour subscription quota each used.
    Finding
    GPT-6 Luna was the quickest and lightest of the four at 2 min 18 s and 1% of the five-hour quota, but the author dismissed its result as a flop (“拉完了”), ranking it alongside MiMo-V2.6-Flash.
    Limitation
    One prompt judged by the author without a score; the quota shares are a subscription meter, not API cost. X's automatic translation renders the verdict as its opposite, so read the original Chinese.

  • Four models build a lava lamp from one prompt

    By Bridgebench (@bridgebench). BridgeMind's benchmark account, which calls itself the official AI benchmark of the vibe coding movement Published 2026-09-23; retrieved 2026-09-24; verified 2026-09-24.

    Task
    Build a lava lamp from a single prompt
    Method
    Same prompt on Claude Opus 5.5, GPT-6 Astra, GPT-6 Sol and GPT-6 Luna; the account reported cost and time, asked readers which did best and commented on their answers in replies.
    Finding
    GPT-6 Luna cost under $0.01 and took 43 seconds, the cheapest and fastest of the four (Claude Opus 5.5: $1.27 and 9 min 35 s); the account left the quality verdict to readers.
    Limitation
    One prompt with the quality verdict left mostly to readers; the account is BridgeMind's own benchmark and does not say how the models were accessed.

  • GPT-6 Luna at max reasoning on a personal chip benchmark

    By Iam_ (@SPAC89). X user who follows LLM releases and runs a personal chip benchmark Published 2026-09-23; retrieved 2026-09-24; verified 2026-09-24.

    Task
    The author's chip benchmark
    Method
    Run on GPT-6 Luna at max reasoning in ChatGPT and compared with GPT-6 Astra; the author reported time, usage and how the model worked.
    Finding
    GPT-6 Luna at max reasoning searched for and generated its own images, launched about 15 sub-agents to research and judge the result, and took almost twice as long as Astra while using about 1% of the weekly usage on the ChatGPT x20 plan; the author expects to make it their default model.
    Limitation
    The benchmark and its scoring are not described and no score is given; usage is a subscription meter, not API cost.

  • AI Coding Daily LLM Coding Leaderboard

    By AI Coding Daily (@AICodingDaily). Coding-model testing channel and website that runs its own leaderboard and sells premium tutorials and channel advertising Published 2026-09-24; retrieved 2026-09-24; verified 2026-09-24.

    Task
    Implement and harden production-shaped projects in several stacks, including a Laravel app, a Go service and a React/TypeScript app, from the same prompts
    Method
    Each model and reasoning setting is a separate row run in a coding harness, with the average cost and time per run. Code quality is graded by a judge model against a published 100-point rubric and scaled to 20 points per project; behavioural reliability runs each of four projects five times and scores the checks every attempt fails.
    Finding
    GPT-6 Luna scored 51.99 at max reasoning for about $0.03 and 11 min per run in Codex CLI, 11th of 47 rows and just above GPT-5.6 Luna at max (51.13); at medium reasoning it fell to 40.20.
    Limitation
    One author's benchmark whose code-quality half is graded by a judge model; rows run in different harnesses, the page is updated as models are added (read on 2026-09-24, 47 rows), and the author sells premium tutorials and advertising.

  • kzhu's six-question LLM test leaderboard

    By AI产品狙击手 kzhu (@kevinzhu9305). Chinese YouTube reviewer who scores models on a six-question test and publishes every score with per-question notes Published 2026-09-24; retrieved 2026-09-24; verified 2026-09-24.

    Task
    Six questions scored out of 10: constrained sentence writing, two reasoning puzzles (replaced by pelican and compound-bow SVG animations in version 2.2), a browser OS with a space shooter, a 3D racing game and a Trello-style board
    Method
    Each row names the harness the model ran in (Pi, Claude Code, WorkBuddy and others) and the date; the author scores every question with a note, and warns that totals are only comparable within one test version.
    Finding
    GPT-6 Luna scored 40/60 on version 2.2 in Pi (the author's note says 41): full marks on the constrained sentences, but physically wrong SVG animations (4 and 6), a flat 2D space shooter, a cluttered racing track and small text on the Trello board.
    Limitation
    One author's scores on six questions; rows use different harnesses and test versions (2.1 or 2.2), and a few rows' table total differs by one point from the author's note.