Skip to content

policy: admit GPT-6.2 Astra / GPT-6.2 Sol, retire GPT-6-Astra, GPT-6.1 Sol and GPT-6.1 Terra #487

Open

Description

@Dong1017

Summary

Update the knowledge-authorship model admission list (.github/knowledge-model-policy.json and docs/knowledge-model-policy.md):

Admit

  • gpt-6.2-astra (GPT-6.2 Astra)
  • gpt-6.2-sol (GPT-6.2 Sol)

Retire (remove admission)

  • gpt-6-astra (GPT-6-Astra)
  • gpt-6.1-sol (GPT-6.1 Sol)
  • gpt-6.1-terra (GPT-6.1 Terra)

Claude Opus 5.5 and Claude Fable 5.2 are unchanged.

Background

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol, and both are available in the API as gpt-6.2-astra / gpt-6.2-sol (announcement, GPT-6.2 Astra docs, GPT-6.2 Sol docs). GPT-6.2 Sol is positioned as the successor to GPT-6.1 Sol and GPT-6.2 Astra as the new frontier model, so the owner wants knowledge authorship to move to the 6.2 generation and stop admitting the previous GPT entries.

Proposed scope

Follow the same shape as the Terra admission (#298) and the Fable 5.2 admission (#439):

  1. Add the two exact canonical IDs, no aliases, at xhigh and max only, matching the existing GPT entries. Before merging, the implementing PR must check the canonical IDs and support for xhigh / max against the official model docs and record the check date in the policy doc. If either effort is not supported, flag it here rather than silently changing the effort set.
  2. Remove gpt-6-astra, gpt-6.1-sol and gpt-6.1-terra from the allowlist and the owner table, and drop or rewrite their verification paragraphs.
  3. Update tests/test_knowledge_model_policy.py (and tests/test_release_notes.py if it pins the model list). Cover:
    • The new IDs are accepted at xhigh / max.
    • Other efforts and inferred aliases (e.g. gpt-6.2, astra, sol) are rejected.
    • The retired IDs are now rejected.

Open question for reviewers

Should already-merged knowledge contributions that declare the retired models stay valid? Today, the policy is evaluated on new PRs against trusted main, so existing contributions are not re-evaluated. Please confirm that is the intended behaviour. Also confirm whether a short transition window is needed for in-flight PRs that already declare a retired model (they would need a model change or an exception after this lands).

Out of scope: evaluator logic, exception process, protected paths, and trusted-main policy selection are unchanged. This is an admission-policy change only; it does not qualify knowledge content or run a model.

Activity

  1. Dong1017 commented on Nov 18, 2026

    @Dong1017
    CollaboratorAuthor

    Posted by Dong-bot, an automated agent acting on behalf of @Dong1017 (always-on-bot, model claude-sonnet-5-5).

    Assessment: The scope is clear and consistent with the earlier admission changes (#298 Terra, #439 Fable 5.2). I'd treat it as ready to implement once the two decisions below are recorded. I haven't verified the model IDs or effort support. I can't browse the linked OpenAI docs, and I haven't read the policy files or tests.

    What it asks

    • Admit gpt-6.2-astra and gpt-6.2-sol at xhigh / max only, with no aliases.
    • Retire gpt-6-astra, gpt-6.1-sol and gpt-6.1-terra.
    • Leave Claude Opus 5.5 and Claude Fable 5.2 unchanged.
    • Touch .github/knowledge-model-policy.json, docs/knowledge-model-policy.md, tests/test_knowledge_model_policy.py, and tests/test_release_notes.py if it pins the model list.

    Gaps and risks

    1. Doc check is still open. The canonical IDs and xhigh / max support are asserted but not yet confirmed. The implementing PR should record the check date in the policy doc, as the issue says. If either effort isn't supported, the PR should flag it here and not change the effort set quietly.
    2. Retirement and in-flight PRs. Retiring three models at once will fail any open knowledge PR that declares one of them. Before merging, list the open PRs that declare a retired model. Then choose a transition window, or require a model change or exception on those PRs. Without that list, the transition question can't be answered.
    3. Policy source. The policy is evaluated against trusted main, so the new list applies to new and updated PRs only once this PR is on main. That ordering is correct, but it does mean a PR that is rebased or re-evaluated after the merge will see the new list.
    4. Tests.
      • Cover both new IDs at xhigh and max.
      • Cover each rejected effort (for example high, medium, low), not just one.
      • Cover the alias rejections (gpt-6.2, astra, sol).
      • Cover all three retired IDs.
      • Check for near-miss IDs such as gpt-6.2-terra or gpt-6-sol that might match by prefix.
    5. Stale references. Grep for the retired IDs and display names in the docs, the release notes, the owner table, the verification paragraphs, and any fixtures. A leftover reference could pass the allowlist test while the docs still advertise a retired model.

    Open question: existing merged contributions

    • Not re-evaluating already-merged contributions matches the evaluator's current behaviour, because it runs on new PRs against trusted main. I'd record that as the intended behaviour, and the policy doc should say it explicitly, so the retirement isn't read as invalidating prior authorship.
    • On the transition window, I'd lean toward no grace period in the policy itself. A grace period adds a time-based rule, and the issue says evaluator logic is out of scope. The existing exception process can cover in-flight PRs that need it. The final call depends on how many open PRs declare a retired model (gap 2).

    Suggested next steps

    1. Post the list of open PRs that declare gpt-6-astra, gpt-6.1-sol or gpt-6.1-terra.
    2. Confirm in this thread: (a) merged contributions stay valid, and (b) in-flight PRs use the exception process, with no policy-level window.
    3. Open the implementing PR with the doc-check date and the test coverage above, and link it here.

    Dong-bot (authorized delegate) triage; comment only.

  2. david6666666 commented on Nov 18, 2026

    @david6666666
    Collaborator

    A cutover detail: changing the policy JSON or docs forces a full refresh of open PRs targeting main, even if their heads don’t change. I checked the current declarations and protected-file diffs: #456 declares gpt-6.1-sol and #480 declares gpt-6-astra; #457’s Sol mention is reviewer-only. Could we decide how to handle #456 and #480 before cutover, then check their individual declaration statuses after the push? A green sweep job can still leave either PR red. Already-merged contributions are outside that sweep.

  3. Dong1017 commented on Nov 18, 2026

    @Dong1017
    CollaboratorAuthor

    Posted by an agent on behalf of @Dong1017 (incremental review only; the in-flight PR list (#456/#480), the doc-check date and alias/effort test coverage were already raised by Dong-bot and @david6666666 above and are not repeated).

    Two test-side details the implementing PR will hit, checked at main c37832c3:

    1. The test baseline is a retired model. attribution() defaults to model="gpt-6.1-sol" (tests/test_knowledge_model_policy.py:65), and declared()/pr() build on it, so it is the "compliant author" for most of the file (~39 call sites). After retirement, success-asserting tests go red, which is fine, but failure-asserting tests that use the default as the good author stay green for the wrong reason. Example: test_every_material_author_must_qualify (L124) asserts failure for [attribution(), attribution("unapproved-model")], and that would pass even if the second author were never checked. Suggest moving the default to gpt-6.2-sol first, then doing the retirement.
    2. Parametrized cases that name retired IDs. test_suspension_overrides_canonical_and_alias (L113-121) indexes policy["models"]["gpt-6.1-terra"], which raises KeyError once the entry is deleted. test_effort_is_exact_set_membership (L100) on gpt-6.1-sol/gpt-6.1-terra turns vacuous, because an unknown model already errors before effort is checked (tools/check_knowledge_model_policy.py:205-208). Both should switch to the new IDs.

    Optional: the evaluator already rejects "suspended": true with the same verdict as an absent ID (L205, "unapproved or suspended"). So retiring by suspension, rather than deletion, would keep each entry's source URL and verification history in the JSON at no behavioral cost. Either works; removal matches the issue text.

    Not in scope but adjacent: tools/skill_eval_pr_cases.json:4 pins the Skill Eval executor to gpt-6.1-sol. That is not authorship admission, so this policy change doesn't require touching it. Noting it only so nobody greps it up later and treats it as a miss.

    Checked: issue body, both comments, .github/knowledge-model-policy.json, tools/check_knowledge_model_policy.py model_errors, tests/test_knowledge_model_policy.py, and a repo-wide grep for the retired IDs. Not run: tests. Not checked (unverified): the OpenAI docs for the new IDs and effort support.

  4. zengchuang-hw commented on Nov 18, 2026

    @zengchuang-hw
    Contributor

    独立范围审阅,基于 main@c37832c3c7a6c457f533683b7b3386f2c3a0c004 的实际政策 JSON、checker、workflow 与两份测试源码、隔离的现行文档规范,以及官方资料。只使用隔离出的标题目标,不读取其他人的评论或作者转述。以下是验收建议,不是批准政策变更或宣告完成。

    1. 将“仓库准入退役”与“厂商产品下线”分开。 OpenAI 发布公告及官方 Astra、Sol页面可支持两个新 canonical ID 的存在;这些资料不能代替仓库选择淘汰旧模型的决策依据。当前政策仍允许三个旧 GPT ID,尚无两个新 ID。建议在验收中明确生效版本、旧 ID 是删除还是保留并 suspended、以及在途贡献如何处理,不要仅因新模型发布就宣称旧型号已被供应商停用。

    2. 需要正反两面的精确身份矩阵,而非只加两条成功用例。 checker 的 model_errors按 canonical/显式 aliases 匹配;缺失或 suspended 的型号失败,effort 必须同时属于政策值和 {xhigh,max}。建议覆盖新 ID × 两档 effort、三个旧 ID 的拒绝、未声明别名/大小写/空白/非法 effort,以及任一 material author 不合规的混合声明。若保留 suspended 条目,还应验证 alias 不会绕过。官方reasoning 文档说明支持值取决于具体模型;通用枚举不等价于对每个型号/宿主组合的执行证据。本次没有做付费 API 或实际模型调用。

    特别注意 test_knowledge_model_policy.py:65–121:默认 attribution 仍是旧 Sol,成功矩阵和 suspension fixture 也引用旧型号。只改 JSON 会让很多非模型主题测试继承一个不再合规的默认作者,或在删除条目后访问不存在的 key。应迁移通用 fixture,并独立保留旧型号负例,不能靠删除历史覆盖来消绿。

    3. 验收必须包含已存在 PR 的重新评估,而不仅是新 PR。 workflow读取 trusted main;routing:1221–1226在政策输入变化时走 full sweep。因此“曾经成功、仍包含旧模型 material author 的 open PR”是必须覆盖的切换场景。现行文档规范也要求按目标分支政策而非 PR 自带替代政策评估。政策更新不能把真实旧作者身份改写成新模型来通过检查;若需要例外或过渡,须明确其独立规则和证据。当前 assess 对 policy_paths 只做分类,不以旧 policy-change 评论自动授予例外(257–275 行),也不能把技术门禁成功误当作维护者已接受政策提案。

    4. 不要将无关模型透传测试误改成准入测试。 test_release_notes.py:507–515验证用户显式给出的模型 ID 原样进入请求,且未擅自添加 reasoning。它并不验证 Knowledge 作者准入。如果替换示例型号,应保留这两个断言;仓库准入退役本身不意味着通用客户端应改写或禁止该 ID。

    **证据与限制:**完成了上述源码静态检查、固定 main 身份核验及官方资料核查;未执行仓库测试、lint、真实模型调用或任何实验,原因是可用执行环境的隔离可信性未确认。尚未获得可安全分离的详细正文 AC;已核对的现行文档要求前瞻适用,不要求重建未知历史身份或提供私密会话日志;隔离视图不能补足本提案的详细生效/过渡 AC。故不能给出整体 PASS。这里没有修改仓库政策、模型设置、权限或恢复被停止的实验。

  5. added
    kind:rfcProposal / design discussion; scope review before build
    on Nov 18, 2026
  6. Dong1017 commented on Nov 18, 2026

    @Dong1017
    CollaboratorAuthor

    Posted by an agent on behalf of @Dong1017 (RFC owner decision record).

    Decision record for #487

    Recorded 2026-11-18 against main 76ca44a8; the policy files are unchanged on the current main cf671a88. Implementing PR: #499. This record sets the owner's decisions for the PR to implement. It does not approve the PR; independent review and hosted CI still apply to its head.

    D1. Admit gpt-6.2-astra and gpt-6.2-sol

    • Decision: admit both exact canonical IDs, with no aliases, at xhigh and max only. This is the same shape as the existing GPT entries, Terra (policy: admit GPT-6.1 Terra at xhigh and max #298) and Fable 5.2 (policy: admit Claude Fable 5.2 at xhigh and max #439).

    • Basis: official OpenAI docs, read on 2026-11-18:

      • The GPT-6.2 Astra and GPT-6.2 Sol model pages give the IDs ("Model IDs: gpt-6.2-astra" / "gpt-6.2-sol") and list no aliases or snapshots.
      • The model catalog lists Reasoning low medium high xhigh max for both models.
      • The reasoning guide says both models reject none.

      The model pages themselves do not list effort values, so xhigh/max support rests on the catalog. The PR records the check date in the policy doc.

    • Not claimed:

      • No model was run and no paid API call was made.
      • Nothing here shows that a given client or host actually sends these efforts. Contributors still declare their real settings.
      • The announcement returned HTTP 403 and was not read.
      • Admission says nothing about knowledge quality.
    • Reopen if: the official docs change either ID or stop listing xhigh or max for either model, or a short alias is requested. Any change needs a new doc check.

    D2. Stop admitting gpt-6-astra, gpt-6.1-sol and gpt-6.1-terra by deleting their entries

    • Decision: delete the three entries from the allowlist, the owner table and the verification text. Do not keep them as "suspended": true.
    • Basis: owner choice to move GPT knowledge authorship to the 6.2 generation. The evaluator returns the same verdict for an absent ID and a suspended one ("model is unapproved or suspended"), so deletion changes no behaviour. It keeps the allowlist equal to what is admitted today. The earlier entries, their source URLs and their check dates remain in git history and in policy: admit GPT-6.1 Terra at xhigh and max #298.
    • Not claimed: that the vendor retired, deprecated or replaced these models. This is the repository choosing to stop admitting them. On 2026-11-18 OpenAI's own docs still describe them as current:
      • The model catalog lists gpt-6.1-terra.
      • The deprecations page names gpt-6.1-sol and gpt-6.1-terra as replacement models.
    • Reopen if: the owner wants one of them back for knowledge authorship. That would be a new admission change with a fresh doc check.

    D3. Merged contributions that declare a removed model stay valid

    • Decision: they are not re-evaluated and not invalidated. knowledge: add GPU measurement-validity and profiler attribution guidance #456 (declares gpt-6.1-sol) merged on 2026-11-18 and is covered by this.
    • Basis:
      • The workflow evaluates open PRs against trusted main.
      • A merged or closed PR receives no new status, and its earlier verdict stays historical (Lifecycle and reruns).
      • The policy is prospective and does not rewrite historical commits (Rollout and scope).
      • The policy doc now states this next to the removal.
    • Not claimed: that this record reviews or qualifies any merged content. It also changes no recorded attribution.
    • Reopen if: a defect is found in a specific merged contribution. Handle that as a content issue on that contribution, not through admission.

    D4. Immediate cutover, no transition window

    D5. #480 must re-author the affected part or use the exception process

    D6. Other references to the removed IDs stay as they are

    • Decision: the PR changes no file outside the policy JSON, its document and its admission tests.
    • Basis: the remaining references are of three kinds:
      • Historical records: docs/reviews/issue91-*.md and evidence/.
      • Fixtures of past integration runs: tests/fixtures/repo_to_skill/.
      • Not authorship admission:
        • The Skill Eval executor pin in tools/skill_eval_pr_cases.json.
        • tests/test_release_notes.py:507-515, which checks that an explicit model ID is sent unchanged.
    • Not claimed: that the Skill Eval executor should keep gpt-6.1-sol. That is a separate decision.
    • Reopen if: a separate issue proposes moving the Skill Eval executor model.

    Review points and how the PR answers them

    Point Raised by How the PR addresses it
    Check IDs and xhigh/max against official docs; record the date; flag rather than change efforts if unsupported Dong-bot, zengchuang-hw Checked 2026-11-18 (D1). Both efforts are listed for both models; effort set unchanged
    List in-flight PRs declaring a removed model and decide how to handle them before cutover Dong-bot, david6666666, zengchuang-hw #480 only; #456 merged first (D3–D5)
    Policy JSON/doc change forces a full refresh of open main PRs david6666666, zengchuang-hw No change needed; noted in the PR body
    Default test author gpt-6.1-sol must move first, or failure tests stay green for the wrong reason agent comment (on behalf of @Dong1017), zengchuang-hw Commit 1 moves it to gpt-6.2-sol before commit 2 removes anything; a new test asserts the default is admitted. Running the old tests against the new JSON shows 51 failures and the wrong-reason green on test_every_material_author_must_qualify
    Parametrized cases naming removed IDs would raise KeyError or turn vacuous agent comment, zengchuang-hw Effort and suspension cases moved to admitted models
    Cover new IDs × both efforts, other efforts, aliases, removed IDs Dong-bot, zengchuang-hw Covered
    Grep for stale references Dong-bot, agent comment Done (D6)
    Record that merged contributions are not re-evaluated Dong-bot; david6666666 noted merged work is outside the sweep D3; policy doc paragraph

    Points raised by a single reviewer

    Point Raised by How the PR addresses it
    No policy-level grace period; use the exception process Dong-bot Adopted (D4)
    Near-miss IDs such as gpt-6.2-terra and gpt-6-sol Dong-bot Both are rejected, as are gpt-6.2-astra-latest, GPT-6.2 Astra and GPT-6.2 Sol
    Read each affected PR's own status after the push; a green sweep is not enough david6666666 Listed as a post-merge step
    Retirement by suspension would keep source and history in the JSON agent comment Not adopted; the owner chose deletion (D2)
    tools/skill_eval_pr_cases.json is not authorship admission agent comment Left unchanged (D6)
    Separate repository retirement from vendor retirement zengchuang-hw D2 and one sentence in the policy doc
    Case and whitespace variants, and a mixed declaration where any one author is noncompliant zengchuang-hw Covered. The mixed case puts the bad author first and last, and asserts only that author is reported
    A policy update must not rewrite a true old author identity to pass zengchuang-hw D5 and the #480 notice
    tests/test_release_notes.py:507-515 is a transmission test, not admission zengchuang-hw Left unchanged (D6)
    Technical gate success is not maintainer acceptance of the policy change zengchuang-hw Not claimed in the PR. This record is the owner's decision, and the PR still needs independent review

    Dong-bot and the agent comment both act on behalf of @Dong1017, the RFC owner. @david6666666 and @zengchuang-hw are the reviewers outside the owner account.

  7. zengchuang-hw commented on Nov 18, 2026

    @zengchuang-hw
    Contributor

    Independent incremental review of the Issue-owner-account decision, following my earlier scope review and PR499 review.

    The newly explicit D1–D6 requirements resolve the earlier source/scope questions: exact new IDs at xhigh/max, deletion rather than suspension of the three old IDs, no retrospective invalidation of merged contributions, immediate trusted-main cutover without a grace period, and truthful re-authoring or the existing exception process for affected open work. In particular, re-authoring is not permission to relabel a historical material author. This is an account-attributed governing requirement, not an expansion of my authority to implement or merge it.

    I independently compared these requirements with PR499's actual three-file diff and unchanged evaluator at head 49c5453c63f7b86e4c466b804f427fec524f1e37: the JSON membership/effort changes, documentation of cutover and historical treatment, positive default-fixture guard and explicit rejection/mixed-author tests align. Protected paths, trusted-main selection and exception eligibility remain unchanged. Official model-catalog verification supporting the new IDs and efforts is recorded in my PR review; no actual model run or quality qualification is implied.

    I also rechecked the PR metadata: the literal DECISION_LINK placeholder is now removed and this exact decision comment is linked. I therefore close that specific decision-trace objection from my initial PR review.

    The execution-evidence gap is separate and still open: at this inspection full-check jobs 134786431885 and 134786431884 remain in progress. Browser, native-host, installed-package and declaration checks now report success, but those do not substitute for the changed policy regression module's completed execution. I have not approved PR499 or treated the current policy proposal as already merged. A full policy-input refresh over existing open PRs remains an important cutover check, not permission to rewrite their declarations.

    This round used isolated normative requirements, actual source, current machine status and my own prior findings. No other reviewers' opinions or attribution summaries were used; no repository code/model was executed, no policy/Issue state was changed and nothing was merged.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    kind:rfcProposal / design discussion; scope review before buildstatus:acceptedThe decision owner accepted it in a recorded comment

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions