BrowseComp 130-question subset — Mercor web research agent Effort: Max Configuration: Mercor web research agent, 100 steps, 3 hours 62.3% mercor.com for DeepSeek-V4-Pro (max) on BrowseComp 130-question subset — Mercor web research agent (opens in a new tab) Archived copy (opens in a new tab) BrowseComp Accuracy Reported October 7, 2026 Benchmark leaderboard Mercor Independent Record for DeepSeek-V4-Pro on BrowseComp 130-question subset — Mercor web research agent BrowseComp Effort: High 80.4% huggingface.co for DeepSeek-V4-Pro (high) on BrowseComp BrowseComp (opens in a new tab) BrowseComp Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on BrowseComp BrowseComp BrowseComp Effort: Max 83.4% huggingface.co for DeepSeek-V4-Pro (max) on BrowseComp BrowseComp (opens in a new tab) BrowseComp Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on BrowseComp BrowseComp DeepSWE 1.1 Effort: Max 63.0% deepswe.datacurve.ai for DeepSeek-V4-Pro (max) on DeepSWE 1.1 (opens in a new tab) DeepSWE Pass@1 Reported September 22, 2026 Benchmark leaderboard Datacurve Independent Record for DeepSeek-V4-Pro on DeepSWE 1.1 DeepSWE 1.1 — Mercor mini-swe-agent, 500 steps Effort: Max Configuration: Mercor mini-swe-agent, 500 steps, 2 hours 9.1% mercor.com for DeepSeek-V4-Pro (max) on DeepSWE 1.1 — Mercor mini-swe-agent, 500 steps (opens in a new tab) Archived copy (opens in a new tab) DeepSWE Pass@1 Reported October 7, 2026 Benchmark leaderboard Mercor Independent Record for DeepSeek-V4-Pro on DeepSWE 1.1 — Mercor mini-swe-agent, 500 steps Finance Agent Benchmark 2 — October 2026 evaluation Effort: Max 44.1% vals.ai for DeepSeek-V4-Pro (max) on Finance Agent Benchmark 2 — October 2026 evaluation (opens in a new tab) Archived copy (opens in a new tab) Finance Agent Accuracy Reported October 6, 2026 Benchmark leaderboard Vals AI Independent Record for DeepSeek-V4-Pro on Finance Agent Benchmark 2 — October 2026 evaluation GPQA Diamond Effort: None Reported as: non-thinking 72.9% huggingface.co for DeepSeek-V4-Pro (non-thinking) on Graduate-Level Google-Proof Q&A Diamond (opens in a new tab) GPQA Diamond Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on Graduate-Level Google-Proof Q&A Diamond GPQA Diamond Effort: High 89.1% huggingface.co for DeepSeek-V4-Pro (high) on Graduate-Level Google-Proof Q&A Diamond (opens in a new tab) GPQA Diamond Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on Graduate-Level Google-Proof Q&A Diamond GPQA Diamond Effort: Max 90.1% huggingface.co for DeepSeek-V4-Pro (max) on Graduate-Level Google-Proof Q&A Diamond (opens in a new tab) GPQA Diamond Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on Graduate-Level Google-Proof Q&A Diamond GPQA Diamond — Epoch AI harness Effort: None Configuration: Epoch AI harness 73.2% epoch.ai for DeepSeek-V4-Pro (none) on Graduate-Level Google-Proof Q&A Diamond — Epoch AI harness (opens in a new tab) Archived copy (opens in a new tab) GPQA Diamond Accuracy Reported October 7, 2026 Evaluated August 6, 2026 at 11:58 PM UTC Benchmark leaderboard Epoch AI Independent Record for DeepSeek-V4-Pro on Graduate-Level Google-Proof Q&A Diamond — Epoch AI harness GPQA Diamond — Epoch AI harness Effort: High Configuration: Epoch AI harness 90.9% epoch.ai for DeepSeek-V4-Pro (high) on Graduate-Level Google-Proof Q&A Diamond — Epoch AI harness (opens in a new tab) Archived copy (opens in a new tab) GPQA Diamond Accuracy Reported October 7, 2026 Evaluated August 6, 2026 at 11:58 PM UTC Benchmark leaderboard Epoch AI Independent Record for DeepSeek-V4-Pro on Graduate-Level Google-Proof Q&A Diamond — Epoch AI harness GPQA Diamond — Epoch AI harness Effort: Max Configuration: Epoch AI harness 89.6% epoch.ai for DeepSeek-V4-Pro (max) on Graduate-Level Google-Proof Q&A Diamond — Epoch AI harness (opens in a new tab) Archived copy (opens in a new tab) GPQA Diamond Accuracy Reported October 7, 2026 Evaluated June 16, 2026 at 8:16 PM UTC Benchmark leaderboard Epoch AI Independent Record for DeepSeek-V4-Pro on Graduate-Level Google-Proof Q&A Diamond — Epoch AI harness GPQA Diamond — Mercor single-shot Effort: Max Configuration: Mercor single-shot agent 89.4% mercor.com for DeepSeek-V4-Pro (max) on Graduate-Level Google-Proof Q&A Diamond — Mercor single-shot (opens in a new tab) Archived copy (opens in a new tab) GPQA Diamond Accuracy Reported October 7, 2026 Benchmark leaderboard Mercor Independent Record for DeepSeek-V4-Pro on Graduate-Level Google-Proof Q&A Diamond — Mercor single-shot GPQA Diamond — Vals AI harness Effort: Max Configuration: Vals AI harness: mean of zero-shot and few-shot CoT 89.4% vals.ai for DeepSeek-V4-Pro (max) on Graduate-Level Google-Proof Q&A Diamond — Vals AI harness (opens in a new tab) Archived copy (opens in a new tab) GPQA Diamond Accuracy Reported September 1, 2026 Benchmark leaderboard Vals AI Independent Record for DeepSeek-V4-Pro on Graduate-Level Google-Proof Q&A Diamond — Vals AI harness HLAB October 2026 evaluation Effort: Max 3.8% vals.ai for DeepSeek-V4-Pro (max) on Harvey's Legal Agent Benchmark October 2026 evaluation (opens in a new tab) Archived copy (opens in a new tab) HLAB Task Pass Rate Reported October 6, 2026 Benchmark leaderboard Vals AI Independent Record for DeepSeek-V4-Pro on Harvey's Legal Agent Benchmark October 2026 evaluation Humanity's Last Exam Full set — No tools Effort: Max Configuration: No tools 37.7% huggingface.co for DeepSeek-V4-Pro (max) on Humanity's Last Exam Full set — No tools (opens in a new tab) Humanity's Last Exam Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on Humanity's Last Exam Full set — No tools Humanity's Last Exam Full set — With tools Effort: Max Configuration: With tools 48.2% huggingface.co for DeepSeek-V4-Pro (max) on Humanity's Last Exam Full set — With tools (opens in a new tab) Humanity's Last Exam Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on Humanity's Last Exam Full set — With tools Humanity's Last Exam Text only — Mercor single-shot Effort: Max Configuration: Mercor single-shot agent 34.4% mercor.com for DeepSeek-V4-Pro (max) on Humanity's Last Exam Text only — Mercor single-shot (opens in a new tab) Archived copy (opens in a new tab) Humanity's Last Exam Accuracy Reported October 7, 2026 Benchmark leaderboard Mercor Independent Record for DeepSeek-V4-Pro on Humanity's Last Exam Text only — Mercor single-shot MMLU-Pro Original Effort: None Reported as: non-thinking 82.9% huggingface.co for DeepSeek-V4-Pro (non-thinking) on MMLU-Pro Original (opens in a new tab) Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on MMLU-Pro Original MMLU-Pro Original Effort: High 87.1% huggingface.co for DeepSeek-V4-Pro (high) on MMLU-Pro Original (opens in a new tab) Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on MMLU-Pro Original MMLU-Pro Original Effort: Max 87.5% huggingface.co for DeepSeek-V4-Pro (max) on MMLU-Pro Original (opens in a new tab) Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on MMLU-Pro Original MMLU-Pro Original — Mercor single-shot Effort: Max Configuration: Mercor single-shot agent 86.5% mercor.com for DeepSeek-V4-Pro (max) on MMLU-Pro Original — Mercor single-shot (opens in a new tab) Archived copy (opens in a new tab) Accuracy Reported October 7, 2026 Benchmark leaderboard Mercor Independent Record for DeepSeek-V4-Pro on MMLU-Pro Original — Mercor single-shot MMLU-Pro Original — Vals AI harness Effort: Max Configuration: Vals AI harness 87.2% vals.ai for DeepSeek-V4-Pro (max) on MMLU-Pro Original — Vals AI harness (opens in a new tab) Archived copy (opens in a new tab) Accuracy Reported September 1, 2026 Benchmark leaderboard Vals AI Independent Record for DeepSeek-V4-Pro on MMLU-Pro Original — Vals AI harness Riemann-bench October 2026 leaderboard Effort: Max 38.4% surgehq.ai for DeepSeek-V4-Pro (Max) on Riemann-bench October 2026 leaderboard (opens in a new tab) Archived copy (opens in a new tab) Riemann-bench Pass@1 Reported October 7, 2026 Benchmark leaderboard Surge AI Independent Record for DeepSeek-V4-Pro on Riemann-bench October 2026 leaderboard SimpleQA Verified Original Effort: None Reported as: non-thinking 45.0% huggingface.co for DeepSeek-V4-Pro (non-thinking) on SimpleQA Verified Original (opens in a new tab) Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on SimpleQA Verified Original SimpleQA Verified Original Effort: High 46.2% huggingface.co for DeepSeek-V4-Pro (high) on SimpleQA Verified Original (opens in a new tab) Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on SimpleQA Verified Original SimpleQA Verified Original Effort: Max 57.9% huggingface.co for DeepSeek-V4-Pro (max) on SimpleQA Verified Original (opens in a new tab) Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on SimpleQA Verified Original SimpleQA Verified Original — Epoch AI harness Effort: Max Configuration: Epoch AI harness 47.0% epoch.ai for DeepSeek-V4-Pro (max) on SimpleQA Verified Original — Epoch AI harness (opens in a new tab) Archived copy (opens in a new tab) Accuracy Reported October 7, 2026 Evaluated August 27, 2026 at 7:30 PM UTC Benchmark leaderboard Epoch AI Independent Record for DeepSeek-V4-Pro on SimpleQA Verified Original — Epoch AI harness SWE Atlas Codebase QnA — Mercor mini-swe-agent, 250 steps Effort: Max Configuration: Mercor mini-swe-agent, 250 steps, 1 hour 25.0% mercor.com for DeepSeek-V4-Pro (max) on SWE Atlas Codebase QnA — Mercor mini-swe-agent, 250 steps (opens in a new tab) Archived copy (opens in a new tab) Mean pass@1 Reported October 7, 2026 Benchmark leaderboard Mercor Independent Record for DeepSeek-V4-Pro on SWE Atlas Codebase QnA — Mercor mini-swe-agent, 250 steps SWE Atlas Codebase QnA — Public; mini-swe-agent Configuration: mini-swe-agent harness 27.2% labs.scale.com for DeepSeek-V4-Pro on SWE Atlas Codebase QnA — Public; mini-swe-agent (opens in a new tab) Archived copy (opens in a new tab) Mean pass@1 Reported October 7, 2026 Benchmark leaderboard Scale AI Independent Record for DeepSeek-V4-Pro on SWE Atlas Codebase QnA — Public; mini-swe-agent SWE-bench Multilingual Effort: None Reported as: non-thinking 69.8% huggingface.co for DeepSeek-V4-Pro (non-thinking) on SWE-bench Multilingual (opens in a new tab) Resolved Tasks Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on SWE-bench Multilingual SWE-bench Multilingual Effort: High 74.1% huggingface.co for DeepSeek-V4-Pro (high) on SWE-bench Multilingual (opens in a new tab) Resolved Tasks Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on SWE-bench Multilingual SWE-bench Multilingual Effort: Max 76.2% huggingface.co for DeepSeek-V4-Pro (max) on SWE-bench Multilingual (opens in a new tab) Resolved Tasks Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on SWE-bench Multilingual SWE-bench Verified Effort: None Reported as: non-thinking 73.6% huggingface.co for DeepSeek-V4-Pro (non-thinking) on SWE-bench Verified (opens in a new tab) Resolved Tasks Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on SWE-bench Verified SWE-bench Verified Effort: High 79.4% huggingface.co for DeepSeek-V4-Pro (high) on SWE-bench Verified (opens in a new tab) Resolved Tasks Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on SWE-bench Verified SWE-bench Verified Effort: Max 80.6% huggingface.co for DeepSeek-V4-Pro (max) on SWE-bench Verified (opens in a new tab) Resolved Tasks Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on SWE-bench Verified SWE-bench Verified — Epoch AI harness Effort: Max Configuration: Epoch AI harness 77.6% epoch.ai for DeepSeek-V4-Pro (max) on SWE-bench Verified — Epoch AI harness (opens in a new tab) Archived copy (opens in a new tab) Resolved Tasks Reported October 7, 2026 Evaluated June 18, 2026 at 5:43 PM UTC Benchmark leaderboard Epoch AI Independent Record for DeepSeek-V4-Pro on SWE-bench Verified — Epoch AI harness SWE-bench Verified — Mercor mini-swe-agent, 1000 steps Effort: Max Configuration: Mercor mini-swe-agent, 1000 steps, 3 hours 75.4% mercor.com for DeepSeek-V4-Pro (max) on SWE-bench Verified — Mercor mini-swe-agent, 1000 steps (opens in a new tab) Archived copy (opens in a new tab) Resolved Tasks Reported October 7, 2026 Benchmark leaderboard Mercor Independent Record for DeepSeek-V4-Pro on SWE-bench Verified — Mercor mini-swe-agent, 1000 steps SWE-bench Verified — Vals AI mini-swe-agent Effort: Max Configuration: mini-swe-agent harness 77.4% vals.ai for DeepSeek-V4-Pro (max) on SWE-bench Verified — Vals AI mini-swe-agent (opens in a new tab) Archived copy (opens in a new tab) Resolved Tasks Reported September 1, 2026 Benchmark leaderboard Vals AI Independent Record for DeepSeek-V4-Pro on SWE-bench Verified — Vals AI mini-swe-agent SWE-bench Pro Public Effort: Max 55.4% huggingface.co for DeepSeek-V4-Pro (max) on SWE-bench Pro Public (opens in a new tab) Public tasks resolved Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on SWE-bench Pro Public Terminal-Bench 2.0 Effort: None Reported as: non-thinking 59.1% huggingface.co for DeepSeek-V4-Pro (non-thinking) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on Terminal-Bench 2.0 Terminal-Bench 2.0 Effort: High 63.3% huggingface.co for DeepSeek-V4-Pro (high) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on Terminal-Bench 2.0 Terminal-Bench 2.0 Effort: Max 67.9% huggingface.co for DeepSeek-V4-Pro (max) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on Terminal-Bench 2.0 Terminal-Bench 2.0 — Vals AI Terminus 2 Configuration: Terminus-2 harness 56.2% vals.ai for DeepSeek-V4-Pro on Terminal-Bench 2.0 — Vals AI Terminus 2 (opens in a new tab) Archived copy (opens in a new tab) Terminal-Bench Accuracy Reported June 4, 2026 Benchmark leaderboard Vals AI Independent Record for DeepSeek-V4-Pro on Terminal-Bench 2.0 — Vals AI Terminus 2 Terminal-Bench 2.1 — Mercor Terminus 2 Effort: Max Configuration: Mercor Terminus 2, 1000 steps, 3 hours 66.7% mercor.com for DeepSeek-V4-Pro (max) on Terminal-Bench 2.1 — Mercor Terminus 2 (opens in a new tab) Archived copy (opens in a new tab) Terminal-Bench Accuracy Reported October 7, 2026 Benchmark leaderboard Mercor Independent Record for DeepSeek-V4-Pro on Terminal-Bench 2.1 — Mercor Terminus 2 Terminal-Bench 2.1 — Vals AI Terminus 2 Configuration: Terminus-2 harness 50.2% vals.ai for DeepSeek-V4-Pro on Terminal-Bench 2.1 — Vals AI Terminus 2 (opens in a new tab) Archived copy (opens in a new tab) Terminal-Bench Accuracy Reported September 28, 2026 Benchmark leaderboard Vals AI Independent Record for DeepSeek-V4-Pro on Terminal-Bench 2.1 — Vals AI Terminus 2 Terminal-Bench 4.0 — Vals AI mini-swe-agent Effort: Max Configuration: mini-swe-agent harness 11.1% vals.ai for DeepSeek-V4-Pro (max) on Terminal-Bench 4.0 — Vals AI mini-swe-agent (opens in a new tab) Archived copy (opens in a new tab) Terminal-Bench Accuracy Reported October 6, 2026 Benchmark leaderboard Vals AI Independent Record for DeepSeek-V4-Pro on Terminal-Bench 4.0 — Vals AI mini-swe-agent Toolathlon Original Effort: None Reported as: non-thinking 46.3% huggingface.co for DeepSeek-V4-Pro (non-thinking) on Toolathlon Original (opens in a new tab) Tasks completed Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on Toolathlon Original Toolathlon Original Effort: High 49.0% huggingface.co for DeepSeek-V4-Pro (high) on Toolathlon Original (opens in a new tab) Tasks completed Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on Toolathlon Original Toolathlon Original Effort: Max 51.8% huggingface.co for DeepSeek-V4-Pro (max) on Toolathlon Original (opens in a new tab) Tasks completed Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on Toolathlon Original