Benchmark Almanac

IFEvalCode

Deepseek-R1 (snapshot unspecified) 43.1% tops its leaderboard Claude 3.7 Sonnet 38.2% leads the models we track
Not in model cards

Models write code in eight programming languages under explicit constraints. The evaluator checks functional correctness and adherence to instructions separately.

Made by
Jian Yang and colleagues
Released
July 2025
Kind
Function-level problems
Level
University
Scored
Checked automatically
Answers
Code
Tasks
All public
Measures
Capability
Language
Chinese, English
In and out
Text, Code → Code
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
Strongest model tested
Claude 3.7 Sonnet #119 on the capability index
Last model added
April 2025
Scores refreshed
November 20th, 2026
42
Entries in its paper, 25 models we track

Deepseek-R1 (snapshot unspecified) tops the leaderboard with 43.1% Claude 3.7 Sonnet leads with 38.2% among models released by April 2025

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The paper reports separate averages by prompt language. The headline is the plain mean of English and Chinese code correctness. 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 Claude 3.7 Sonnet, at 38.2%. This view shows only models we track, each at its best.

The newest model with a score here came out in April 2025, 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-R1 (snapshot unspecified) 43.1%
  2. 2 ControlledCoder-32B 41.2%
  3. 3 Claude 3.7 Sonnet 38.2%
  4. 4 GPT-4.1 36.6%
  5. 5 Deepseek-V3 (snapshot unspecified) 34.5%
  6. 6 GPT-4.1 mini 34.4%
  7. 7 GPT-4o (snapshot unspecified) 32.9%
  8. 8 Claude 3.5 Sonnet 32.5%
  9. 9 Qwen3-30B-A3B Non-thinking 31.3%
  10. 10 Qwen3-32B Thinking 30.6%
  11. 11 Qwen3-235B-A22B Thinking 30.4%
  12. 12 Qwen2.5-Coder-32B 29.8%
  13. 13 Qwen2.5-Coder-14B 29.3%
  14. 14 GPT-4o-mini (snapshot unspecified) 28.6%
  15. 15 Llama 3.1-70B 27.9%
  16. 16 Qwen3-14B Thinking 27%
  17. 17 DeepSeek Coder 33B 26.8%
  18. 18 Yi-Coder-9B 26.6%
  19. 19 Qwen2.5-Coder (7B) 25.5%
  20. 20 grok-3-mini-fast 25.4%
  21. 21 OpenCoder-8B 22.3%
  22. 22 DeepSeek Coder 6.7B 21.8%
  23. 23 CodeQwen1.5-7B 20.2%
  24. 24 Grok 3 19.6%
  25. 25 OpenCoder-1.5B 17.9%
  26. 26 Llama 3.1-8B 17.6%
  27. 27 Granite-Coder-34B 17.5%
  28. 28 Qwen3-8B Thinking 17.5%
  29. 29 Qwen2.5-Coder-3B 17.4%
  30. 30 Granite-Coder-8B 16.4%
  31. 31 Yi-Coder-1.5B 16.3%
  32. 32 Granite-Coder-20B 16.2%
  33. 33 Qwen3-4B Non-thinking 15.1%
  34. 34 Qwen2.5-Coder (1.5B) 12.7%
  35. 35 Granite-Coder-3B 12.6%
  36. 36 CodeLlama-13B 12.3%
  37. 37 Qwen3-1.7B Non-thinking 11.8%
  38. 38 CodeLlama-34B 11.3%
  39. 39 DeepSeek Coder 1.3B 10.5%
  40. 40 Qwen2.5-Coder-0.5B 8.65%
  41. 41 CodeLlama-7B 7.35%
  42. 42 Qwen3-0.6B Non-thinking 5.40%

Source: The IFEvalCode paper

#EntryScoreSettingDateSource
1 Deepseek-R1 (snapshot unspecified) 43.1% Official leaderboard
2 ControlledCoder-32B 41.2% Official leaderboard
3 Claude 3.7 Sonnet 38.2% Feb 2025 Official leaderboard
4 GPT-4.1 36.6% Apr 2025 Official leaderboard
5 Deepseek-V3 (snapshot unspecified) 34.5% Official leaderboard
6 GPT-4.1 mini 34.4% Apr 2025 Official leaderboard
7 GPT-4o (snapshot unspecified) 32.9% Official leaderboard
8 Claude 3.5 Sonnet 32.5% Jun 2024 Official leaderboard
9 Qwen3-30B-A3B 31.3% Non-thinking Apr 2025 Official leaderboard
10 Qwen3-32B 30.6% Thinking Apr 2025 Official leaderboard
11 Qwen3-235B-A22B 30.4% Thinking Apr 2025 Official leaderboard
12 Qwen2.5-Coder-32B 29.8% Sep 2024 Official leaderboard
13 Qwen2.5-Coder-14B 29.3% Sep 2024 Official leaderboard
14 GPT-4o-mini (snapshot unspecified) 28.6% Official leaderboard
15 Llama 3.1-70B 27.9% Jul 2024 Official leaderboard
16 Qwen3-14B 27% Thinking Apr 2025 Official leaderboard
17 DeepSeek Coder 33B 26.8% Nov 2023 Official leaderboard
18 Yi-Coder-9B 26.6% Official leaderboard
19 Qwen2.5-Coder (7B) 25.5% Sep 2024 Official leaderboard
20 grok-3-mini-fast 25.4% Official leaderboard
21 OpenCoder-8B 22.3% Official leaderboard
22 DeepSeek Coder 6.7B 21.8% Nov 2023 Official leaderboard
23 CodeQwen1.5-7B 20.2% Apr 2024 Official leaderboard
24 Grok 3 19.6% Apr 2025 Official leaderboard
25 OpenCoder-1.5B 17.9% Official leaderboard
26 Llama 3.1-8B 17.6% Jul 2024 Official leaderboard
27 Granite-Coder-34B 17.5% Official leaderboard
28 Qwen3-8B 17.5% Thinking Apr 2025 Official leaderboard
29 Qwen2.5-Coder-3B 17.4% Sep 2024 Official leaderboard
30 Granite-Coder-8B 16.4% Official leaderboard
31 Yi-Coder-1.5B 16.3% Official leaderboard
32 Granite-Coder-20B 16.2% Official leaderboard
33 Qwen3-4B 15.1% Non-thinking Apr 2025 Official leaderboard
34 Qwen2.5-Coder (1.5B) 12.7% Sep 2024 Official leaderboard
35 Granite-Coder-3B 12.6% Official leaderboard
36 CodeLlama-13B 12.3% Official leaderboard
37 Qwen3-1.7B 11.8% Non-thinking Apr 2025 Official leaderboard
38 CodeLlama-34B 11.3% Official leaderboard
39 DeepSeek Coder 1.3B 10.5% Nov 2023 Official leaderboard
40 Qwen2.5-Coder-0.5B 8.65% Sep 2024 Official leaderboard
41 CodeLlama-7B 7.35% Official leaderboard
42 Qwen3-0.6B 5.40% Non-thinking Apr 2025 Official leaderboard
#ModelScoreSettingDateSource
1 Claude 3.7 Sonnet 38.2% Feb 2025 Official leaderboard
2 GPT-4.1 36.6% Apr 2025 Official leaderboard
3 GPT-4.1 mini 34.4% Apr 2025 Official leaderboard
4 Claude 3.5 Sonnet 32.5% Jun 2024 Official leaderboard
5 Qwen3-30B-A3B 31.3% Non-thinking Apr 2025 Official leaderboard
6 Qwen3-32B 30.6% Thinking Apr 2025 Official leaderboard
7 Qwen3-235B-A22B 30.4% Thinking Apr 2025 Official leaderboard
8 Qwen2.5-Coder-32B 29.8% Sep 2024 Official leaderboard
9 Qwen2.5-Coder-14B 29.3% Sep 2024 Official leaderboard
10 Llama 3.1-70B 27.9% Jul 2024 Official leaderboard
11 Qwen3-14B 27% Thinking Apr 2025 Official leaderboard
12 DeepSeek Coder 33B 26.8% Nov 2023 Official leaderboard
13 Qwen2.5-Coder (7B) 25.5% Sep 2024 Official leaderboard
14 DeepSeek Coder 6.7B 21.8% Nov 2023 Official leaderboard
15 CodeQwen1.5-7B 20.2% Apr 2024 Official leaderboard
16 Grok 3 19.6% Apr 2025 Official leaderboard
17 Llama 3.1-8B 17.6% Jul 2024 Official leaderboard
18 Qwen3-8B 17.5% Thinking Apr 2025 Official leaderboard
19 Qwen2.5-Coder-3B 17.4% Sep 2024 Official leaderboard
20 Qwen3-4B 15.1% Non-thinking Apr 2025 Official leaderboard
21 Qwen2.5-Coder (1.5B) 12.7% Sep 2024 Official leaderboard
22 Qwen3-1.7B 11.8% Non-thinking Apr 2025 Official leaderboard
23 DeepSeek Coder 1.3B 10.5% Nov 2023 Official leaderboard
24 Qwen2.5-Coder-0.5B 8.65% Sep 2024 Official leaderboard
25 Qwen3-0.6B 5.40% Non-thinking Apr 2025 Official leaderboard

Rankings by part

We read 36 parts from the benchmark’s paper, each ranking models on one part of the test, such as English Python Correctness, English Python Instruction adherence and English Java Correctness. Claude 3.7 Sonnet leads 8 of them, and 11 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

English Python Correctness

25 models we track have a score here.

  1. 1 Qwen3-30B-A3B ↑4 54.3%
  2. 2 Qwen3-14B ↑9 51.4%
  3. 3 GPT-4.1 ↓1 50.5%
  4. 4 Qwen3-235B-A22B ↑3 50.5%
  5. 5 Qwen3-32B ↑1 49.5%
  6. 6 Claude 3.5 Sonnet ↓2 48.6%
  7. 7 Qwen2.5-Coder-14B ↑2 48.6%
  8. 8 Claude 3.7 Sonnet ↓7 47.6%
  9. 9 GPT-4.1 mini ↓6 46.7%
  10. 10 Llama 3.1-70B 42.9%

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
Qwen3-30B-A3B
Thinking Non-thinking
31.3%
Qwen3-32B
Non-thinking Thinking
30.6%
Qwen3-235B-A22B
Non-thinking
30.4%
Qwen3-14B
Non-thinking
27%
Qwen3-8B
Non-thinking
17.5%
Qwen3-4B
Thinking
15.1%
Qwen3-1.7B
Thinking
11.8%
Qwen3-0.6B
Thinking
5.40%

Source: The IFEvalCode paper

ModelEffortHarnessScoreCost per taskOutput tokens
Claude 3.7 Sonnet 38.2%
GPT-4.1 36.6%
GPT-4.1 mini 34.4%
Claude 3.5 Sonnet 32.5%
Qwen3-30B-A3B Non-thinking 31.3%
Qwen3-32B Thinking 30.6%
Qwen3-235B-A22B Thinking 30.4%
Qwen2.5-Coder-32B 29.8%
Qwen2.5-Coder-14B 29.3%
Llama 3.1-70B 27.9%
Qwen3-235B-A22B Non-thinking 27.8%
Qwen3-32B Non-thinking 27.6%
Qwen3-14B Thinking 27%
DeepSeek Coder 33B 26.8%
Qwen3-30B-A3B Thinking 26.4%
Qwen3-14B Non-thinking 25.8%
Qwen2.5-Coder (7B) 25.5%
DeepSeek Coder 6.7B 21.8%
CodeQwen1.5-7B 20.2%
Grok 3 19.6%
Llama 3.1-8B 17.6%
Qwen3-8B Thinking 17.5%
Qwen2.5-Coder-3B 17.4%
Qwen3-8B Non-thinking 16.7%
Qwen3-4B Non-thinking 15.1%
Qwen3-4B Thinking 14.1%
Qwen2.5-Coder (1.5B) 12.7%
Qwen3-1.7B Non-thinking 11.8%
Qwen3-1.7B Thinking 10.8%
DeepSeek Coder 1.3B 10.5%
Qwen2.5-Coder-0.5B 8.65%
Qwen3-0.6B Non-thinking 5.40%
Qwen3-0.6B Thinking 4.30%

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
Qwen3-30B-A3B
Thinking Non-thinking
31.3%
Qwen3-32B
Non-thinking Thinking
30.6%
Qwen3-235B-A22B
Non-thinking
30.4%
Qwen3-14B
Non-thinking
27%
Qwen3-8B
Non-thinking
17.5%
Qwen3-4B
Thinking
15.1%
Qwen3-1.7B
Thinking
11.8%
Qwen3-0.6B
Thinking
5.40%

Source: The IFEvalCode paper

ModelEffortHarnessScoreCost per taskOutput tokens
Claude 3.7 Sonnet 38.2%
GPT-4.1 36.6%
GPT-4.1 mini 34.4%
Claude 3.5 Sonnet 32.5%
Qwen3-30B-A3B Non-thinking 31.3%
Qwen3-32B Thinking 30.6%
Qwen3-235B-A22B Thinking 30.4%
Qwen2.5-Coder-32B 29.8%
Qwen2.5-Coder-14B 29.3%
Llama 3.1-70B 27.9%
Qwen3-235B-A22B Non-thinking 27.8%
Qwen3-32B Non-thinking 27.6%
Qwen3-14B Thinking 27%
DeepSeek Coder 33B 26.8%
Qwen3-30B-A3B Thinking 26.4%
Qwen3-14B Non-thinking 25.8%
Qwen2.5-Coder (7B) 25.5%
DeepSeek Coder 6.7B 21.8%
CodeQwen1.5-7B 20.2%
Grok 3 19.6%
Llama 3.1-8B 17.6%
Qwen3-8B Thinking 17.5%
Qwen2.5-Coder-3B 17.4%
Qwen3-8B Non-thinking 16.7%
Qwen3-4B Non-thinking 15.1%
Qwen3-4B Thinking 14.1%
Qwen2.5-Coder (1.5B) 12.7%
Qwen3-1.7B Non-thinking 11.8%
Qwen3-1.7B Thinking 10.8%
DeepSeek Coder 1.3B 10.5%
Qwen2.5-Coder-0.5B 8.65%
Qwen3-0.6B Non-thinking 5.40%
Qwen3-0.6B Thinking 4.30%

Number one over time

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

Anthropic
Benchmark paper Not tested since Jul 2025

Source: The IFEvalCode 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 Best so far
0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) Claude 3.7 Sonnet Claude 3.5 Sonnet DeepSeek Coder 33B 0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) Claude 3.7 Sonnet DeepSeek Coder 33B

Source: Official IFEvalCode 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

9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 19 months ago, in April 2025, for Qwen3-0.6B. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
  • Pass Independent runs Scores from the benchmark’s paper (42 entries).
  • Pass Models scored 25 models we track, plus 17 other entries on its leaderboard.
  • Warning Newest score The newest came 19 months ago, in April 2025, for Qwen3-0.6B.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Claude 3.7 Sonnet (#119).
  • Pass Headroom The best is 38.2%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 33 rows for models we track and 17 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