Benchmark Almanac

TSP-Opt

o3-mini 56.4% leads since March 2025
o3-mini 56.4% leads since March 2025
Only in system cards

Models find optimal tours for 960 traveling salesman instances across 8 difficulty levels. Answers are checked for optimality.

Microsoft released the dataset on Hugging Face as microsoft/tsp in June 2025, and the Eureka nphard_tsp config loads it.

Released
March 2025
Kind
Abstract puzzles and logic
Level
University
Scored
Checked automatically
Answers
Short answer
Strongest model tested
o1 #117 on the capability index
Last in a system card
April 2025
Last model added
April 2025
Scores refreshed
November 12th, 2026
14
Entries on its own leaderboard, 11 models we track
2
System cards report it, from 1 lab

o3-mini leads with 56.4% among models released by February 2025

Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

o3-mini leads with 56.4%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model.

The newest model with a score here came out in February 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

14 entries. Scroll sideways for the rest.

  1. 56.4 o3-mini
  2. 47.7 Claude 3.7 Sonnet
  3. 46.6 DeepSeek-R1
  4. 45.8 o1
  5. 42.6 vllm-local-reasoning-phi-rl_checkpoints/phi-15b-rl
  6. 39.6 vllm-local-reasoning-phi-sft_14b_allmathv2_code_2e_coco_5e_high32k_10xlr
  7. 37.5 vllm-local-reasoning-phi-sft_16b_phi4_mathv2code5_otcode_oaif_safe_high32_2
  8. 32.4 DeepSeek-R1-Distill-Llama-70B
  9. 30.1 Gemini 2.0 Flash Thinking
  10. 20.5 Gemini 2.0 Pro
  11. 13.3 Claude 3.5 Sonnet (October 2024)
  12. 10.2 GPT-4o
  13. 9.22 Llama 3.1-405B
  14. 8.15 Phi-4
  1. 1 o3-mini mean across runs; high; OpenAI/nphard_tsp_o3_mini_high_1/2025-03-07-11-08-50.020866 56.4%
  2. 2 Claude 3.7 Sonnet mean across runs; thinking budget 60000; Claude/nphard_tsp_CLAUDE_3_7_SONNET_THINKING/2025-03-20-16-15-16.404019 47.7%
  3. 3 DeepSeek-R1 mean across runs; DeepSeek/nphard_tsp1_deepseek_together/2025-03-14-13-36-39.101455 46.6%
  4. 4 o1 mean across runs; OpenAI/nphard_tsp_o1/2025-02-11-11-08-42.764778 45.8%
  5. 5 vllm-local-reasoning-phi-rl_checkpoints/phi-15b-rl mean across runs; Phi/PHI_4_REASONING_phi-15b-rl/2025-04-17-07-30-20.949372 42.6%
  6. 6 vllm-local-reasoning-phi-sft_14b_allmathv2_code_2e_coco_5e_high32k_10xlr mean across runs; Phi/Phi-4_reasoning/2025-04-09-23-23-45.057159 39.6%
  7. 7 vllm-local-reasoning-phi-sft_16b_phi4_mathv2code5_otcode_oaif_safe_high32_2 mean across runs; Phi/PHI_4_REASONING_phi-16b/2025-04-16-21-36-06.239655 37.5%
  8. 8 DeepSeek-R1-Distill-Llama-70B mean across runs; DeepSeek/DeepSeek-r1-Distill-Llama-70B/2025-04-10-00-47-05.953934 32.4%
  9. 9 Gemini 2.0 Flash Thinking Jan 2025, mean across runs; Gemini/nphard_gemini-2.0-flash-thinking-exp-0121_no_code/2025-03-17-13-18-19.917609 30.1%
  10. 10 Gemini 2.0 Pro mean across runs; Gemini/nphard_gemini-2.0-pro/2025-03-05-08-38-27.381311 20.5%
  11. 11 Claude 3.5 Sonnet (October 2024) mean across runs; Claude/nphard_tsp_level_claude-3-5-sonnet-20241022/2025-02-11-11-00-24.996719 13.3%
  12. 12 GPT-4o Aug 2024, mean across runs; OpenAI/nphard_tsp_gpt4o_2024_08_06/2025-03-13-16-48-24.776136 10.2%
  13. 13 Llama 3.1-405B mean across runs; LLaMA/nphard_tsp_level_llama_3_1_2024-11-20_2/2025-03-17-18-13-20.604359 9.22%
  14. 14 Phi-4 mean across runs; Phi/Phi-4_temp08/2025-04-10-00-53-28.446901 8.15%

Source: Official TSP-Opt leaderboard

13 models. Scroll sideways for the rest.

  1. 56.4 o3-mini
  2. 47.7 Claude 3.7 Sonnet
  3. 46.6 DeepSeek-R1
  4. 45.8 o1
  5. 42.6 Phi-4-Reasoning-plus
  6. 37.5 Phi-4-reasoning
  7. 32.4 DeepSeek-R1-Distill-Llama-70B
  8. 30.1 Gemini 2.0 Flash Thinking
  9. 20.5 Gemini 2.0 Pro
  10. 13.3 Claude 3.5 Sonnet (October 2024)
  11. 10.2 GPT-4o
  12. 9.22 Llama 3.1-405B
  13. 8.15 Phi-4
  1. 1 o3-mini mean across runs; high; OpenAI/nphard_tsp_o3_mini_high_1/2025-03-07-11-08-50.020866 56.4%
  2. 2 Claude 3.7 Sonnet mean across runs; thinking budget 60000; Claude/nphard_tsp_CLAUDE_3_7_SONNET_THINKING/2025-03-20-16-15-16.404019 47.7%
  3. 3 DeepSeek-R1 mean across runs; DeepSeek/nphard_tsp1_deepseek_together/2025-03-14-13-36-39.101455 46.6%
  4. 4 o1 mean across runs; OpenAI/nphard_tsp_o1/2025-02-11-11-08-42.764778 45.8%
  5. 5 Phi-4-Reasoning-plus 42.6%
  6. 6 Phi-4-reasoning 37.5%
  7. 7 DeepSeek-R1-Distill-Llama-70B mean across runs; DeepSeek/DeepSeek-r1-Distill-Llama-70B/2025-04-10-00-47-05.953934 32.4%
  8. 8 Gemini 2.0 Flash Thinking Jan 2025, mean across runs; Gemini/nphard_gemini-2.0-flash-thinking-exp-0121_no_code/2025-03-17-13-18-19.917609 30.1%
  9. 9 Gemini 2.0 Pro mean across runs; Gemini/nphard_gemini-2.0-pro/2025-03-05-08-38-27.381311 20.5%
  10. 10 Claude 3.5 Sonnet (October 2024) mean across runs; Claude/nphard_tsp_level_claude-3-5-sonnet-20241022/2025-02-11-11-00-24.996719 13.3%
  11. 11 GPT-4o Aug 2024, mean across runs; OpenAI/nphard_tsp_gpt4o_2024_08_06/2025-03-13-16-48-24.776136 10.2%
  12. 12 Llama 3.1-405B mean across runs; LLaMA/nphard_tsp_level_llama_3_1_2024-11-20_2/2025-03-17-18-13-20.604359 9.22%
  13. 13 Phi-4 mean across runs; Phi/Phi-4_temp08/2025-04-10-00-53-28.446901 8.15%

Sources: Official TSP-Opt leaderboard and labs’ model cards

#EntryScoreSettingDateSource
1 o3-mini 56.4% mean across runs; high; OpenAI/nphard_tsp_o3_mini_high_1/2025-03-07-11-08-50.020866 Jan 2025 Official leaderboard
2 Claude 3.7 Sonnet 47.7% mean across runs; thinking budget 60000; Claude/nphard_tsp_CLAUDE_3_7_SONNET_THINKING/2025-03-20-16-15-16.404019 Feb 2025 Official leaderboard
3 DeepSeek-R1 46.6% mean across runs; DeepSeek/nphard_tsp1_deepseek_together/2025-03-14-13-36-39.101455 Jan 2025 Official leaderboard
4 o1 45.8% mean across runs; OpenAI/nphard_tsp_o1/2025-02-11-11-08-42.764778 Dec 2024 Official leaderboard
5 vllm-local-reasoning-phi-rl_checkpoints/phi-15b-rl 42.6% mean across runs; Phi/PHI_4_REASONING_phi-15b-rl/2025-04-17-07-30-20.949372 Official leaderboard
6 vllm-local-reasoning-phi-sft_14b_allmathv2_code_2e_coco_5e_high32k_10xlr 39.6% mean across runs; Phi/Phi-4_reasoning/2025-04-09-23-23-45.057159 Official leaderboard
7 vllm-local-reasoning-phi-sft_16b_phi4_mathv2code5_otcode_oaif_safe_high32_2 37.5% mean across runs; Phi/PHI_4_REASONING_phi-16b/2025-04-16-21-36-06.239655 Official leaderboard
8 DeepSeek-R1-Distill-Llama-70B 32.4% mean across runs; DeepSeek/DeepSeek-r1-Distill-Llama-70B/2025-04-10-00-47-05.953934 Jan 2025 Official leaderboard
9 Gemini 2.0 Flash Thinking Jan 2025 30.1% mean across runs; Gemini/nphard_gemini-2.0-flash-thinking-exp-0121_no_code/2025-03-17-13-18-19.917609 Jan 2025 Official leaderboard
10 Gemini 2.0 Pro 20.5% mean across runs; Gemini/nphard_gemini-2.0-pro/2025-03-05-08-38-27.381311 Feb 2025 Official leaderboard
11 Claude 3.5 Sonnet (October 2024) 13.3% mean across runs; Claude/nphard_tsp_level_claude-3-5-sonnet-20241022/2025-02-11-11-00-24.996719 Oct 2024 Official leaderboard
12 GPT-4o Aug 2024 10.2% mean across runs; OpenAI/nphard_tsp_gpt4o_2024_08_06/2025-03-13-16-48-24.776136 Aug 2024 Official leaderboard
13 Llama 3.1-405B 9.22% mean across runs; LLaMA/nphard_tsp_level_llama_3_1_2024-11-20_2/2025-03-17-18-13-20.604359 Jul 2024 Official leaderboard
14 Phi-4 8.15% mean across runs; Phi/Phi-4_temp08/2025-04-10-00-53-28.446901 Dec 2024 Official leaderboard
#ModelScoreSettingDateSource
1 o3-mini 56.4% mean across runs; high; OpenAI/nphard_tsp_o3_mini_high_1/2025-03-07-11-08-50.020866 Jan 2025 Official leaderboard
2 Claude 3.7 Sonnet 47.7% mean across runs; thinking budget 60000; Claude/nphard_tsp_CLAUDE_3_7_SONNET_THINKING/2025-03-20-16-15-16.404019 Feb 2025 Official leaderboard
3 DeepSeek-R1 46.6% mean across runs; DeepSeek/nphard_tsp1_deepseek_together/2025-03-14-13-36-39.101455 Jan 2025 Official leaderboard
4 o1 45.8% mean across runs; OpenAI/nphard_tsp_o1/2025-02-11-11-08-42.764778 Dec 2024 Official leaderboard
5 Phi-4-Reasoning-plus 42.6% Apr 2025 Lab model card
6 Phi-4-reasoning 37.5% Apr 2025 Lab model card
7 DeepSeek-R1-Distill-Llama-70B 32.4% mean across runs; DeepSeek/DeepSeek-r1-Distill-Llama-70B/2025-04-10-00-47-05.953934 Jan 2025 Official leaderboard
8 Gemini 2.0 Flash Thinking Jan 2025 30.1% mean across runs; Gemini/nphard_gemini-2.0-flash-thinking-exp-0121_no_code/2025-03-17-13-18-19.917609 Jan 2025 Official leaderboard
9 Gemini 2.0 Pro 20.5% mean across runs; Gemini/nphard_gemini-2.0-pro/2025-03-05-08-38-27.381311 Feb 2025 Official leaderboard
10 Claude 3.5 Sonnet (October 2024) 13.3% mean across runs; Claude/nphard_tsp_level_claude-3-5-sonnet-20241022/2025-02-11-11-00-24.996719 Oct 2024 Official leaderboard
11 GPT-4o Aug 2024 10.2% mean across runs; OpenAI/nphard_tsp_gpt4o_2024_08_06/2025-03-13-16-48-24.776136 Aug 2024 Official leaderboard
12 Llama 3.1-405B 9.22% mean across runs; LLaMA/nphard_tsp_level_llama_3_1_2024-11-20_2/2025-03-17-18-13-20.604359 Jul 2024 Official leaderboard
13 Phi-4 8.15% mean across runs; Phi/Phi-4_temp08/2025-04-10-00-53-28.446901 Dec 2024 Official leaderboard

Labs report scores for 2 models in their cards. Switch the scores above to rank them together.

Number one over time Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Own leaderboard Not tested since Mar 2025
o3-mini 56.4%

Source: Official TSP-Opt leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) o3-mini DeepSeek-R1 Claude 3.5 Sonnet (October 2024) GPT-4o 0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) o3-mini DeepSeek-R1

Sources: Official TSP-Opt leaderboard and labs’ model cards

Labs report it only in system cards

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.

2 system cards report it, from 1 lab

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Apr 30, 2025 Phi-4-reasoning technical report 37.5 TSP (TSP-Opt), average pass@1 (Figure 8, printed bar label) Capabilities
Apr 30, 2025 Phi-4-Reasoning-plus technical report 42.6 TSP (TSP-Opt), average pass@1 (Figure 8, printed bar label) Capabilities

More in Science and reasoning

All 92 in Science and reasoning
  • GPQA Diamond In use 198 graduate-level 'Google-proof' multiple-choice questions in biology, chemistry and physics. Gemini 4 Carbon 96.5%
  • Terminal-Bench-Science In use 70 scientific research-workflow tasks across five science areas, run in Harbor containers. Hidden tests give the resolution rate. GPT-6 Astra 68.1%
  • FrontierScience In use Models solve physics, chemistry, and biology problems in an Olympiad track graded by short answer and a Research track graded by 10-point rubrics. Intern-S2-397B 87%
  • BioMysteryBench In use Bioinformatics problems on anonymized real datasets that require real analysis with tools. Rubric-graded, with human-solvable and human-difficult splits. Claude Fable 5.2 83.7%
  • HiPhO In use Models solve theory problems from 2024 and 2025 physics olympiad exams, some with diagrams. Official marking schemes grade answers and steps, mapped to medal levels. Gemini 3 Pro 78.3%
  • LABBench2 In use 11 biology research sub-tasks (literature, figures, tables, patents, trials and more) with tool access. Macro-averaged accuracy. GPT-5.2 60.9%