Claim · coding

GLM 5.1on SWE-bench Verified (Epoch's own run)

71.4 to 76.2 the range across every scoreable measurement we hold

4.8 points between the highest and the lowest

73.2 the number the board uses: the lower median, a real row with its own provenance

5 declared configurations, from 3 runners

1 / 4 rows whose cited page still shows this figure, refetched and checked by us

3 independent runners have measured this, which is rare: on this snapshot only 12 of 321 claim pages can say it.

2 of 4 cited pages no longer show the figure we publish. The page was refetched and split at its own structural boundaries, and no block that names this model carries this number. That does not make the measurement wrong: a leaderboard is a living page and a figure recorded when it was there can be gone by the time you follow the link. It does mean the citation, as printed, no longer supports the number, and this is the only place we know of that says so about its own sources.

Every measurement

5 rows. Best first. the board takes the LOWER MEDIAN of the scoreable measurements, which is a real row with its own interval and provenance rather than an interpolated average. a vendor self-report is shown beside the independent numbers and is never one of them; it cannot move a rank.

Score	Provenance	Configuration	Who ran it	Where to check it
76.2 / 100 as published: 76.2%	INDEPENDENT counts toward the ranking	glm-5.1 harness NeMo Evaluator SDK effort model-card defaults variant verified	NVIDIA via Model benchmark dossier facts, not expression	not checked: the page is a format we cannot read2026-08-30 research.nvidia.com
74.2 / 100 as published: 74.2%	INDEPENDENT counts toward the ranking	glm-5.1	Epoch AI 2026-05-15T10:07:59.494Z CC-BY-4.0	NO_URL2026-08-30 benchmark leaderboard the leaderboard for this benchmark, not a link to this row
73.2 / 100 as published: 73.2%	INDEPENDENT counts toward the ranking	glm-5.1 harness GPU baseline; 10 groups over 3 rounds effort not disclosed variant verified	vLLM-Ascend project via Model benchmark dossier 2026-08-15T00:00:00.000Z facts, not expression	checked: found, but not in a block naming only this model2026-08-30"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"[DeepSeek-V4-Flash/GLM-5.1] Intermittent SWE benchmark failures and score variabili github.com
72.0 / 100 as published: 72.0%	INDEPENDENT counts toward the ranking	glm-5.1 harness Ascend A5 W4A4C8 effort not disclosed variant verified	vLLM-Ascend project via Model benchmark dossier 2026-08-15T00:00:00.000Z facts, not expression	checked: found, but not in a block naming only this model2026-08-30"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"[DeepSeek-V4-Flash/GLM-5.1] Intermittent SWE benchmark failures and score variabili github.com
71.4 / 100 as published: 71.4%	INDEPENDENT counts toward the ranking	glm-5.1 harness Ascend A3 dual-node effort not disclosed variant verified	vLLM-Ascend project via Model benchmark dossier 2026-08-15T00:00:00.000Z facts, not expression	checked: this figure appears in the page’s row for this model2026-08-30GLM-5.1: A3 dual-node: 71.4% github.com

← The Forge Rankings

Snapshot 2026-08-30-62e4e85f043b, manifest dc46131315046f1e, as of 2026-08-30. where only one party has measured a model on a benchmark, this page says so rather than presenting a single number as a reconciliation.