Name: Muse Spark 1.2 vs Muse Spark 1.3
Creator: Benchmark Registry

Shared benchmarks

Shared benchmarks for Muse Spark 1.2 and Muse Spark 1.3
Benchmark	Muse Spark 1.2	Muse Spark 1.3
DeepSearchQA 900-prompt set	85.9%	90.3%
Evaluation details for DeepSearchQA 900-prompt set The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry. Muse Spark 1.2 Score 85.9% Evaluator Meta Version DeepSearchQA 900-prompt set Metric F1 score (percent) Reasoning xhigh Reported September 2, 2026 Source Evaluation source for Muse Spark 1.2 on DeepSearchQA 900-prompt set (opens in a new tab) Muse Spark 1.3 Score 90.3% Evaluator Meta Version DeepSearchQA 900-prompt set Metric F1 score (percent) Reasoning max Reported September 2, 2026 Source Evaluation source for Muse Spark 1.3 on DeepSearchQA 900-prompt set (opens in a new tab)
DeepSWE 1.1 — mini-swe-agentPotentially non-equivalent: Evaluator sets differ.	55.0%1.1 — mini-swe-agent · DeepSWE Pass@1 DataCurve	75.4%1.1 — mini-swe-agent · DeepSWE Pass@1 Meta
Potentially non-equivalent · Evaluation details for DeepSWE 1.1 — mini-swe-agent Evaluator sets differ. Evaluation methodology is not recorded by the registry. Muse Spark 1.2 Score 55.0% Evaluator DataCurve Version DeepSWE 1.1 — mini-swe-agent Metric DeepSWE Pass@1 (percent) Reasoning xhigh Reported September 2, 2026 Source Evaluation source for Muse Spark 1.2 on DeepSWE 1.1 — mini-swe-agent (opens in a new tab) Muse Spark 1.3 Score 75.4% Evaluator Meta Version DeepSWE 1.1 — mini-swe-agent Metric DeepSWE Pass@1 (percent) Reasoning max Reported September 2, 2026 Source Evaluation source for Muse Spark 1.3 on DeepSWE 1.1 — mini-swe-agent (opens in a new tab)
JobBench Main 65 tasks	61.6%	64.9%
Evaluation details for JobBench Main 65 tasks The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry. Muse Spark 1.2 Score 61.6% Evaluator Meta Version JobBench Main 65 tasks Metric Mean rubric score (percent) Reasoning xhigh Reported September 2, 2026 Source Evaluation source for Muse Spark 1.2 on JobBench Main 65 tasks (opens in a new tab) Muse Spark 1.3 Score 64.9% Evaluator Meta Version JobBench Main 65 tasks Metric Mean rubric score (percent) Reasoning max Reported September 2, 2026 Source Evaluation source for Muse Spark 1.3 on JobBench Main 65 tasks (opens in a new tab)
MRCR v2 8-needle 256K–512K (o200k_base)	66.3%	98.5%
Evaluation details for MRCR v2 8-needle 256K–512K (o200k_base) The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry. Muse Spark 1.2 Score 66.3% Evaluator Meta Version MRCR v2 8-needle 256K–512K (o200k_base) Metric Mean sequence-matcher ratio (percent) Reasoning xhigh Reported September 2, 2026 Source Evaluation source for Muse Spark 1.2 on MRCR v2 8-needle 256K–512K (o200k_base) (opens in a new tab) Muse Spark 1.3 Score 98.5% Evaluator Meta Version MRCR v2 8-needle 256K–512K (o200k_base) Metric Mean sequence-matcher ratio (percent) Reasoning max Reported September 2, 2026 Source Evaluation source for Muse Spark 1.3 on MRCR v2 8-needle 256K–512K (o200k_base) (opens in a new tab)
MRCR v2 8-needle 512K–1M (o200k_base)	55.5%	98.1%
Evaluation details for MRCR v2 8-needle 512K–1M (o200k_base) The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry. Muse Spark 1.2 Score 55.5% Evaluator Meta Version MRCR v2 8-needle 512K–1M (o200k_base) Metric Mean sequence-matcher ratio (percent) Reasoning xhigh Reported September 2, 2026 Source Evaluation source for Muse Spark 1.2 on MRCR v2 8-needle 512K–1M (o200k_base) (opens in a new tab) Muse Spark 1.3 Score 98.1% Evaluator Meta Version MRCR v2 8-needle 512K–1M (o200k_base) Metric Mean sequence-matcher ratio (percent) Reasoning max Reported September 2, 2026 Source Evaluation source for Muse Spark 1.3 on MRCR v2 8-needle 512K–1M (o200k_base) (opens in a new tab)
SWE Atlas Codebase QnA — Public; mini-swe-agent	46.2%	59.4%
Evaluation details for SWE Atlas Codebase QnA — Public; mini-swe-agent The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry. Muse Spark 1.2 Score 46.2% Evaluator Meta Version SWE Atlas Codebase QnA — Public; mini-swe-agent Metric Mean pass@1 (percent) Reasoning xhigh Reported September 2, 2026 Source Evaluation source for Muse Spark 1.2 on SWE Atlas Codebase QnA — Public; mini-swe-agent (opens in a new tab) Muse Spark 1.3 Score 59.4% Evaluator Meta Version SWE Atlas Codebase QnA — Public; mini-swe-agent Metric Mean pass@1 (percent) Reasoning max Reported September 2, 2026 Source Evaluation source for Muse Spark 1.3 on SWE Atlas Codebase QnA — Public; mini-swe-agent (opens in a new tab)
Terminal-BenchPotentially non-equivalent: Benchmark versions differ.	82.9%2.1 — Muse Code · Terminal-Bench Accuracy Meta	88.8%2.1 · Terminal-Bench Accuracy Meta
Potentially non-equivalent · Evaluation details for Terminal-Bench Benchmark versions differ. Evaluation methodology is not recorded by the registry. Muse Spark 1.2 Score 82.9% Evaluator Meta Version Terminal-Bench 2.1 — Muse Code Metric Terminal-Bench Accuracy (percent) Reasoning xhigh Reported September 2, 2026 Source Evaluation source for Muse Spark 1.2 on Terminal-Bench 2.1 — Muse Code (opens in a new tab) Muse Spark 1.3 Score 88.8% Evaluator Meta Version Terminal-Bench 2.1 Metric Terminal-Bench Accuracy (percent) Reasoning max Reported September 2, 2026 Source Evaluation source for Muse Spark 1.3 on Terminal-Bench 2.1 (opens in a new tab)