Benchmark Almanac

Mercury

deepseek-coder-33b-base 0.66 tops its leaderboard StarCoder 2 15B 0.59 leads the models we track
Not in model cards

Models write Python solutions and pass hidden correctness tests. The evaluator compares runtime with reference solutions. Harbor ports all 256 problems from the evaluation split.

In the 2024 paper, preference-tuned DeepSeek-Coder 33B led Table 2 at 73.4% overall correctness. The paper reports code efficiency separately.

Made by
Mingzhe Du and colleagues
Released
February 2024
Kind
Speed and optimization
Level
University
Scored
Checked automatically
Answers
Code
Tasks
All public
Measures
Efficiency
Language
English
In and out
Text → Code
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
Strongest model tested
StarCoder 2 15B #224 on the capability index
Last model added
April 2024
Scores refreshed
November 20th, 2026
10
Entries in its paper, 4 models we track

deepseek-coder-33b-base tops the leaderboard with 0.66 StarCoder 2 15B leads with 0.59 among models released by April 2024

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is StarCoder 2 15B, at 0.59. This view shows only models we track, each at its best.

The newest model with a score here came out in April 2024, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 deepseek-coder-33b-base DPO 0.66
  2. 2 CodeLlama-34b-hf DPO 0.64
  3. 3 StarCoder 2 15B DPO 0.59
  4. 4 deepseek-coder-6.7b-base DPO 0.50
  5. 5 CodeQwen1.5-7B DPO 0.49
  6. 6 CodeLlama-13b-hf DPO 0.47
  7. 7 StarCoder 2 7B DPO 0.41
  8. 8 deepseek-coder-1.3b-base Original 0.36
  9. 9 StarCoder 2 3B SFT 0.34
  10. 10 CodeLlama-7b-hf DPO 0.33

Source: The Mercury paper

#EntryScoreSettingDateSource
1 deepseek-coder-33b-base 0.66 DPO Official leaderboard
2 CodeLlama-34b-hf 0.64 DPO Official leaderboard
3 StarCoder 2 15B 0.59 DPO Feb 2024 Official leaderboard
4 deepseek-coder-6.7b-base 0.50 DPO Official leaderboard
5 CodeQwen1.5-7B 0.49 DPO Apr 2024 Official leaderboard
6 CodeLlama-13b-hf 0.47 DPO Official leaderboard
7 StarCoder 2 7B 0.41 DPO Feb 2024 Official leaderboard
8 deepseek-coder-1.3b-base 0.36 Original Official leaderboard
9 StarCoder 2 3B 0.34 SFT Feb 2024 Official leaderboard
10 CodeLlama-7b-hf 0.33 DPO Official leaderboard
#ModelScoreSettingDateSource
1 StarCoder 2 15B 0.59 DPO Feb 2024 Official leaderboard
2 CodeQwen1.5-7B 0.49 DPO Apr 2024 Official leaderboard
3 StarCoder 2 7B 0.41 DPO Feb 2024 Official leaderboard
4 StarCoder 2 3B 0.34 SFT Feb 2024 Official leaderboard

Rankings by part

We read 3 parts from the benchmark’s paper: Easy, Medium and Hard. Each ranks models on one part of the test. StarCoder 2 15B leads every one of them. Arrows show how far a model moved from its overall rank among the models we track.

Easy

4 models we track have a score here.

  1. 1 StarCoder 2 15B 0.68
  2. 2 CodeQwen1.5-7B 0.56
  3. 3 StarCoder 2 7B 0.54
  4. 4 StarCoder 2 3B 0.45

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
StarCoder 2 15B
SFT DPO
0.59
CodeQwen1.5-7B
Original
0.49
StarCoder 2 7B
SFT Original
0.41
StarCoder 2 3B
DPO
0.34

Source: The Mercury paper

ModelEffortHarnessScoreCost per taskOutput tokens
StarCoder 2 15B DPO 0.59
StarCoder 2 15B Original 0.49
CodeQwen1.5-7B DPO 0.49
StarCoder 2 15B SFT 0.48
CodeQwen1.5-7B SFT 0.48
CodeQwen1.5-7B Original 0.48
StarCoder 2 7B DPO 0.41
StarCoder 2 7B Original 0.40
StarCoder 2 7B SFT 0.36
StarCoder 2 3B SFT 0.34
StarCoder 2 3B Original 0.33
StarCoder 2 3B DPO 0.32

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
StarCoder 2 15B
SFT DPO
0.59
CodeQwen1.5-7B
Original
0.49
StarCoder 2 7B
SFT Original
0.41
StarCoder 2 3B
DPO
0.34

Source: The Mercury paper

ModelEffortHarnessScoreCost per taskOutput tokens
StarCoder 2 15B DPO 0.59
StarCoder 2 15B Original 0.49
CodeQwen1.5-7B DPO 0.49
StarCoder 2 15B SFT 0.48
CodeQwen1.5-7B SFT 0.48
CodeQwen1.5-7B Original 0.48
StarCoder 2 7B DPO 0.41
StarCoder 2 7B Original 0.40
StarCoder 2 7B SFT 0.36
StarCoder 2 3B SFT 0.34
StarCoder 2 3B Original 0.33
StarCoder 2 3B DPO 0.32

Number one over time

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

Hugging Face
Benchmark paper Not tested since Apr 2024

Source: The Mercury paper

Scores over time

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

Plot with other benchmarks
In the benchmark’s paper
0.40 0.50 0.60 2024 0.40 0.50 0.60 2024

Source: Official Mercury leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Coding

All 523 in Coding
  • SWE-bench Verified In use 500 human-validated GitHub issues from 12 Python repos. Agent writes a patch that must pass hidden tests. Claude Opus 4.7 83.5%
  • LiveCodeBench v6 In use Competitive programming problems collected after model cutoffs from LeetCode, AtCoder, Codeforces. Scored pass@1 on hidden tests. DeepSeek-V4-Pro 93.5%
  • SWE-bench Multilingual In use 300 GitHub-issue tasks across nine programming languages in SWE-bench format. Patch graded by repository tests. Claude Fable 5.2 96.6%
  • DeepSWE In use 113 original long-horizon engineering tasks across 91 repos and five languages, written from scratch. Hand-written verifiers accept alternative solutions. GPT-6 Astra 74.1%
  • SciCode In use Scientist-curated research coding problems split into subproblems across natural sciences. Code checked against test cases. Seed2.1 Pro 59.8%
  • Codeforces rating In use Model solves recent Codeforces contest problems. An Elo-style rating is computed from simulated contest performance. DeepSeek V4.1 Flash 3,471

Data audit

8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 2 years ago, in April 2024, for CodeQwen1.5-7B. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
  • Pass Independent runs Scores from the benchmark’s paper (10 entries).
  • Pass Models scored 4 models we track, plus 6 other entries on its leaderboard.
  • Warning Newest score The newest came 2 years ago, in April 2024, for CodeQwen1.5-7B.
  • Note Frontier models None of the top 10 has a score. The strongest tested is StarCoder 2 15B (#224).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 12 rows for models we track and 6 other entries at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check