Add Kimi K2.7 Code eval results#88
Merged
jerilynzheng merged 2 commits intoJun 17, 2026
Merged
Conversation
Adds the unreleased moonshotai/kimi-k2.7-code model via a patched OpenCode binary (see README §Unreleased models). The model is available through AI Gateway but not yet on models.dev, and requires a temperature=1 override like Kimi K2.6. New experiment configs for both the base and AGENTS.md variants, with display-name and AGENTS_MD_PAIRS entries wired into the export script. Results (agent-eval 0.14.2, evals pinned to next.js@9e4d718): - Base: 18/24 (75%) - + AGENTS.md: 23/24 (96%, +21 pts) - AGENTS.md fixes agent-029, 031, 032, 038, 039; only agent-040 fails both ways agent-029's classification.json was authored by hand: the auto-classifier (Claude Haiku, 200k ctx) exceeded its context on that eval's oversized transcript. Verified a genuine model failure (real transcripts, tests fail on missing "use cache" + revalidateTag), matching agent-032's auto-verdict. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
jerilynzheng
force-pushed
the
jerilyn/add-kimi-k2.7-code
branch
from
960eda9 to
6db0ee1
Compare
Resolve conflicts from GLM 5.2 (vercel#90) landing on main. Both touch the shared agent-results.json and scripts/export-results.ts: - export-results.ts: keep both glm-5.2 and kimi-k2.7-code entries - agent-results.json: union — kimi-k2.7-code (23/24) alongside glm-5.2, all other models preserved from main Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
gaojude
approved these changes
Jun 17, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds the unreleased
moonshotai/kimi-k2.7-codemodel via a patched OpenCode binary (see README §Unreleased models). The model is available through AI Gateway but not yet on models.dev, and requires atemperature: 1override like Kimi K2.6.Changes
kimi-k2.7-codeandkimi-k2.7-code--agents-md, both using a patched OpenCode binary (opencode-linux-x64-kimi-k2.7-code-v1on Blob)AGENTS_MD_PAIRSentries wired intoscripts/export-results.tsagent-results.jsonupdated additively (everything outside the new entry is byte-identical to main)Results
Run under agent-eval 0.14.2, evals pinned to
next.js@9e4d718(matches CI).Note on agent-029 classification
agent-029'sclassification.jsonwas authored by hand: the automatic classifier (Claude Haiku, 200k context) exceeded its context window on this eval's oversized transcript (~207k tokens). Verified a genuine model failure — all 4 runs have real transcripts and fail the assertions for the missing"use cache"directive andrevalidateTag, matching the auto-classification of the siblingagent-032use-cache eval.🤖 Generated with Claude Code