feat: grok-4.6 flagship re-score + grok 1.0.3 epoch (0.0.52) - #51
Merged
Conversation
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
xAI shipped grok-4.6 on 2026-08-12 and the installed grok CLI auto-updated to 1.0.3 with its default model flipped to grok-4.6, so the grok lane has been running 4.6 since then; this release moves every stale surface to match.
routing: the capability table's grok row moves to grok-4.6 at 4/3/5 (AA Intelligence 61 tied with gpt-5.6-sol behind Fable and Opus; Vals SWE-bench Verified 95.6 vs 86.6 for 4.5 at roughly 604s per task vs 182s for sol; terminal-bench v3.0 26 vs 34.6; 500k context unchanged; xhigh effort new and 4.6 only; cached input repriced up to 0.50/M short and 1.00/M long), the grok-4.5 row is removed under the single flagship shape, the sol row's stale SWE-Bench Pro cross reference now cites the Vals pair, both "(it currently lists only grok-4.5)" parentheticals follow the live listing, the five unscored codex SKUs are recorded as deliberately not routed, and the rules manifest is regenerated. the four role grok policy is retained deliberately; an evidence window on 4.6 agentic gains is open before any ownership revisit.
contract epoch: verified-versions moves to grok 1.0.3 after a source spot check of the 2026-08-12 grok-build snapshot found every companion-facing surface unchanged (headless resolvers still drop --no-subagents, no bare --memory flag, --no-auto-update and --no-wait-for-background remain as hidden clap args, snake_case stop reasons, same six sandbox event types with additive optional ProfileApplied fields, version hard gate at all three entrypoints) and live probes covered consult, web, and write on the installed binary. the "verified through" wording family, the README pair, the fake-grok default version string, and the doctor advisory assertion move together. one source caveat recorded in the changelog: GROK_SUBAGENTS_MAX_DEPTH clamps to a floor of 1 and is not a kill switch; GROK_SUBAGENTS=0 and the hard Agent deny remain the effective controls.
verification: full suite green in a full environment (1114 tests, 1113 pass, 0 fail, 1 standing skip). the first live /fusion:smoke wave also ran and passed all five gates on the new default.