Skip to content
Log in

GSM8K history

Snapshots are retained. Added and removed counts compare normalized words with the preceding revision.

r4Seed corpus update
+0 / −0 words

As of 2026-11-04

GSM8K evaluates multi-step grade-school mathematical word-problem solving and, in the original work, the value of ranking generated solutions with a trained verifier.

Revision snapshot

Task and data. GSM8K is an 8,500-problem collection of linguistically varied grade-school arithmetic word problems whose natural-language solutions typically require several calculation steps.

Scoring. The release reports final-answer solve rate. Higher is better when the same sample, candidate count, and answer extraction rules are retained.

What it measures. multi-step grade-school mathematical word-problem solving and, in the original work, the value of ranking generated solutions with a trained verifier.

Comparison rules. Keep the 100-candidate verifier setup separate from single-generation results and modern provider-reported GSM8K variants.

Access and licensing. The Compendium links to the official release and does not reproduce word problems or solutions. Check the dataset repository before reuse.

r3Seed corpus update
+0 / −0 words

As of 2026-07-13

GSM8K evaluates multi-step grade-school mathematical word-problem solving and, in the original work, the value of ranking generated solutions with a trained verifier.

Revision snapshot

Task and data. GSM8K is an 8,500-problem collection of linguistically varied grade-school arithmetic word problems whose natural-language solutions typically require several calculation steps.

Scoring. The release reports final-answer solve rate. Higher is better when the same sample, candidate count, and answer extraction rules are retained.

What it measures. multi-step grade-school mathematical word-problem solving and, in the original work, the value of ranking generated solutions with a trained verifier.

Comparison rules. Keep the 100-candidate verifier setup separate from single-generation results and modern provider-reported GSM8K variants.

Access and licensing. The Compendium links to the official release and does not reproduce word problems or solutions. Check the dataset repository before reuse.

r2Seed corpus update
+0 / −122 words

As of 2026-07-13

GSM8K evaluates multi-step grade-school mathematical word-problem solving and, in the original work, the value of ranking generated solutions with a trained verifier.

Revision snapshot

Task and data. GSM8K is an 8,500-problem collection of linguistically varied grade-school arithmetic word problems whose natural-language solutions typically require several calculation steps.

Scoring. The release reports final-answer solve rate. Higher is better when the same sample, candidate count, and answer extraction rules are retained.

What it measures. multi-step grade-school mathematical word-problem solving and, in the original work, the value of ranking generated solutions with a trained verifier.

Comparison rules. Keep the 100-candidate verifier setup separate from single-generation results and modern provider-reported GSM8K variants.

Access and licensing. The Compendium links to the official release and does not reproduce word problems or solutions. Check the dataset repository before reuse.

r1Seed corpus update

As of 2026-07-13

GSM8K evaluates multi-step grade-school mathematical word-problem solving and, in the original work, the value of ranking generated solutions with a trained verifier. Its published results are preserved here as dated, setup-specific anchors rather than a universal model ranking.

Revision snapshot

Task and data. GSM8K is an 8,500-problem collection of linguistically varied grade-school arithmetic word problems whose natural-language solutions typically require several calculation steps.

Scoring. The release reports final-answer solve rate. Higher is better when the same sample, candidate count, and answer extraction rules are retained.

What it measures. multi-step grade-school mathematical word-problem solving and, in the original work, the value of ranking generated solutions with a trained verifier.

What it does not measure. advanced mathematics, proof validity, broad reasoning, robustness to adversarial wording, tutoring quality, or whether a correct final answer came from valid intermediate reasoning.

Limitations. A solver can reach the correct answer through flawed reasoning; the public static dataset is contamination-prone, and sampling/ranking budgets change performance.

Comparison rules. Keep the 100-candidate verifier setup separate from single-generation results and modern provider-reported GSM8K variants.

Access and licensing. The Compendium links to the official release and does not reproduce word problems or solutions. Check the dataset repository before reuse.

Result-reading note. Every result below was published by the named third party and reviewed by the Compendium on 2026-07-13. Flintglade did not run, reproduce, certify, or independently validate these evaluations. A score only belongs to its recorded benchmark version, split, prompt or scaffold, metric, and subject version; omitted setup details are called out instead of guessed.