AI Benchmark Digest — 2026-06-26

View on AI Benchmark Hub

Daily

New Benchmarks (8)

• SEAL - SWE Atlas - Refactoring (Score): Fable-5 (Claude Code) xHigh leads with 54.76 across 14 models. Scale SEAL SWE Atlas refactoring benchmark measuring software-engineering agents on codebase refactoring tasks.
• DRACO (Score (%)): Claude Mythos 5 leads with 86.4 across 13 models. Perplexity DRACO deep-research agent evaluation reported in Anthropic's Claude Opus 4.8 system card.
• HealthAdminBench (Success Rate (%)): Claude Mythos 5 (browser-use) leads with 51.9 across 11 models. Healthcare administration agent benchmark for prior authorization, appeals, durable medical equipment, payer portals, fax, and EHR-adjacent workflows.
• Vals AI AIME (Accuracy (%)): gemini-3.1-pro-preview leads with 98.12 across 96 models. Vals AI AIME benchmark measuring competition-math problem solving on American Invitational Mathematics Examination-style tasks.
• LLM Stats (APEX-Agents) (Score (%)): Seed 2.1 Pro leads with 33.8 across 5 models.
• LLM Stats (MathArena Apex) (Score (%)): DeepSeek-V4-Pro-Max leads with 90.2 across 6 models.
• LLM Stats (OfficeQA Pro) (Score (%)): Seed 2.1 Pro leads with 72.2 across 5 models.
• LLM Stats (Terminal-Bench 2.1) (Score (%)): Claude Fable 5 leads with 84.3 across 6 models.

New #1 Leaders (18)

• LLM Stats (ZEROBench) (Score (%)): Seed 2.1 Turbo (57.2) beat Muse Spark (33.0) by 24.2.
• SWE-bench Live (Resolved (%)): AMI Agent + Claude-4.6-Oups (63.0) beat SWE-agent + Claude-4.5-Sonnet (40.0) by 23.0.
• LLM Stats (BLINK) (Score (%)): Seed 2.1 Pro (81.4) beat Qwen 3 VL 235B A22B Instruct (70.7) by 10.7.
• Coding Agent Leaderboard - swe-bench-pro--ansible (Score (%)): Opus 4.8 + OpenCode (78.1) beat Opus 4.8 + Claude Code (69.8) by 8.3.
• LLM Stats (OSWorld) (Score (%)): Seed 2.1 Pro (78.8) beat Claude Opus 4.6 (72.7) by 6.1.
• APEX v1 Medicine (MD) (Score (%)): Opus 4.6 (Max) (71.6) beat GPT-5 (66.0) by 5.6.
• LLM Stats (CharXiv-D) (Score (%)): Seed 2.1 Pro (95.5) beat Qwen 3 VL 32B Instruct (90.5) by 5.0.
• LLM Stats (MathVista) (Score (%)): Seed 2.1 Pro (90.7) beat O3 (86.8) by 3.9.
• LLM Stats (BabyVision) (Score (%)): Seed 2.1 Pro (73.7) beat Qwen 3.7 Plus (70.4) by 3.3.
• APEX v1 Consulting (Score (%)): GPT-5.2 Codex (High) (66.9) beat Gemini 3 Flash (64.0) by 2.9.
• Coding Agent Leaderboard (Score (%)): Opus 4.8 + OpenCode (80.8) beat Opus 4.8 + Claude Code (78.3) by 2.5.
• LLM Stats (ERQA) (Score (%)): Seed 2.1 Pro (72.0) beat Qwen 3.7 Plus (69.8) by 2.2.
• LLM Stats (LVBench) (Score (%)): Seed 2.1 Pro (78.0) beat Qwen 3.7 Plus (76.2) by 1.8.
• LLM Stats (MathVision) (Score (%)): Seed 2.1 Pro (94.5) beat Kimi K2.6 (93.2) by 1.3.
• LLM Stats (Video-MME) (Score (%)): Seed 2.1 Pro (89.2) beat Qwen 3.7 Plus (88.0) by 1.2.
• APEX v1 Investment Banking (Score (%)): GPT-5.3 Codex (High) (65.0) beat GPT-5.2 Pro (64.0) by 1.0.
• APEX v1 (Score (%)): GPT-5.4 (High) (67.2) beat GPT-5 (67.0) by 0.2.
• LLM Stats (MCP Atlas) (Score (%)): Seed 2.1 Pro (83.8) beat Gemini 3.5 Flash (83.6) by 0.2.