Benchmark Almanac

ReviewBench

OhMyBug 83% tops its leaderboard Claude Opus 5.5 73.3% leads the models we track
Not in model cards

AI code review systems review 219 real pull requests from 187 repositories in 19 languages. A judge scores found issues against a multi-source golden set for precision, recall and F1.

Research preview released October 5th, 2026. Senior engineers agreed with the ground truth 96.6% of the time. Harbor hosts it as review-bench/reviewbench-public.

Made by
GitHub and Microsoft
Released
October 2026
Kind
Fixing real repositories
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
All public
Strongest model tested
Claude Opus 5.5 #4 on the capability index
Last model added
September 2026
Scores refreshed
November 12th, 2026
17
Entries on its own leaderboard, 4 models we track

OhMyBug tops the leaderboard with 83%

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. The best model we track is Claude Opus 5.5, at 73.3%.

The newest model with a score here came out in September 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

17 entries. Scroll sideways for the rest.

  1. 83.0 OhMyBug
  2. 82.2 intent-review-agent
  3. 75.7 Copilot code review
  4. 73.3 Claude Opus 5.5
  5. 70.2 GPT-5.6 Sol
  6. 68.3 Qodo
  7. 68.2 block/review
  8. 67.7 HeyDeer
  9. 66.1 Devin AI
  10. 65.6 diffox ultra
  11. 64.8 deepseek-flash
  12. 64.0 DeepSeek-V4-Flash
  13. 63.1 Claude Sonnet 5.5
  14. 60.9 Cursor
  15. 59.1 diffox
  16. 35.8 Cubic
  17. 33.4 Greptile
  1. 1 OhMyBug 83%
  2. 2 intent-review-agent 82.2%
  3. 3 Copilot code review high 75.7%
  4. 4 Claude Opus 5.5 73.3%
  5. 5 GPT-5.6 Sol max 70.2%
  6. 6 Qodo 68.3%
  7. 7 block/review panel=sol-opus · gate=off 68.2%
  8. 8 HeyDeer depth=deep · filter=detailed 67.7%
  9. 9 Devin AI 66.1%
  10. 10 diffox ultra 65.6%
  11. 11 deepseek-flash high 64.8%
  12. 12 DeepSeek-V4-Flash 64%
  13. 13 Claude Sonnet 5.5 63.1%
  14. 14 Cursor 60.9%
  15. 15 diffox 59.1%
  16. 16 Cubic 35.8%
  17. 17 Greptile 33.4%

Source: Official ReviewBench leaderboard

#EntryScoreSettingDateSource
1 OhMyBug 83% OhMyBug Oct 2026 Official leaderboard
2 intent-review-agent 82.2% intent-review-agent Oct 2026 Official leaderboard
3 Copilot code review 75.7% high, Copilot code review Oct 2026 Official leaderboard
4 Claude Opus 5.5 73.3% Laude RevEngine Sep 2026 Official leaderboard
5 GPT-5.6 Sol 70.2% max, Codex Jul 2026 Official leaderboard
6 Qodo 68.3% Qodo Sep 2026 Official leaderboard
7 block/review 68.2% panel=sol-opus · gate=off, block/review Oct 2026 Official leaderboard
8 HeyDeer 67.7% depth=deep · filter=detailed, HeyDeer Oct 2026 Official leaderboard
9 Devin AI 66.1% Devin AI Sep 2026 Official leaderboard
10 diffox ultra 65.6% diffox ultra Oct 2026 Official leaderboard
11 deepseek-flash 64.8% high, DeepSeek Harness Reviewer Oct 2026 Official leaderboard
12 DeepSeek-V4-Flash 64% opswiki Apr 2026 Official leaderboard
13 Claude Sonnet 5.5 63.1% Klaussy Sep 2026 Official leaderboard
14 Cursor 60.9% Cursor Sep 2026 Official leaderboard
15 diffox 59.1% diffox Oct 2026 Official leaderboard
16 Cubic 35.8% Cubic Jun 2026 Official leaderboard
17 Greptile 33.4% Greptile Jun 2026 Official leaderboard

Grounded F1 / duration

Each point is one reviewer configuration; highlighted points are not outperformed on both axes by another. The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

40% 60% 80% 100% 02004006008001,000 Duration (s), lower is better Grounded F1 DeepSeek-V4-Flash: Grounded F1 64%, duration 318 s GPT-5.6 Sol (max): Grounded F1 70.2%, duration 568 s GPT-5.6 Sol (ultra): Grounded F1 69.9%, duration 544 s Claude Sonnet 5.5: Grounded F1 63.1%, duration 185 s Claude Sonnet 5.5 (default): Grounded F1 60.2%, duration 176 s Copilot code review (high): Grounded F1 75.7%, duration 974 s. Not in our model list Copilot code review (medium): Grounded F1 72.7%, duration 371 s. Not in our model list Qodo: Grounded F1 68.3%, duration 274 s. Not in our model list block/review (panel=sol-opus · gate=off): Grounded F1 68.2%, duration 197 s. Not in our model list HeyDeer (depth=deep · filter=detailed): Grounded F1 67.7%, duration 242 s. Not in our model list Copilot code review (low): Grounded F1 66.2%, duration 275 s. Not in our model list Devin AI: Grounded F1 66.1%, duration 193 s. Not in our model list diffox ultra: Grounded F1 65.6%, duration 86.8 s. Not in our model list deepseek-flash (high): Grounded F1 64.8%, duration 319 s. Not in our model list Cursor: Grounded F1 60.9%, duration 257 s. Not in our model list diffox: Grounded F1 59.1%, duration 108 s. Not in our model list Cubic: Grounded F1 35.8%, duration 702 s. Not in our model list Greptile: Grounded F1 33.4%, duration 289 s. Not in our model list Claude Opus 5.5: Grounded F1 73.3%, duration 182 s Claude Opus 5.5: Grounded F1 69.3%, duration 55.1 s OhMyBug: Grounded F1 83%, duration 322 s. Not in our model list intent-review-agent: Grounded F1 82.2%, duration 213 s. Not in our model list OhMyBugintent-review-agentClaude Opus 5.5Claude Opus 5.5 40% 60% 80% 100% 02004006008001,000 Duration (s), lower is better Grounded F1 DeepSeek-V4-Flash: Grounded F1 64%, duration 318 s GPT-5.6 Sol (max): Grounded F1 70.2%, duration 568 s GPT-5.6 Sol (ultra): Grounded F1 69.9%, duration 544 s Claude Sonnet 5.5: Grounded F1 63.1%, duration 185 s Claude Sonnet 5.5 (default): Grounded F1 60.2%, duration 176 s Copilot code review (high): Grounded F1 75.7%, duration 974 s. Not in our model list Copilot code review (medium): Grounded F1 72.7%, duration 371 s. Not in our model list Qodo: Grounded F1 68.3%, duration 274 s. Not in our model list block/review (panel=sol-opus · gate=off): Grounded F1 68.2%, duration 197 s. Not in our model list HeyDeer (depth=deep · filter=detailed): Grounded F1 67.7%, duration 242 s. Not in our model list Copilot code review (low): Grounded F1 66.2%, duration 275 s. Not in our model list Devin AI: Grounded F1 66.1%, duration 193 s. Not in our model list diffox ultra: Grounded F1 65.6%, duration 86.8 s. Not in our model list deepseek-flash (high): Grounded F1 64.8%, duration 319 s. Not in our model list Cursor: Grounded F1 60.9%, duration 257 s. Not in our model list diffox: Grounded F1 59.1%, duration 108 s. Not in our model list Cubic: Grounded F1 35.8%, duration 702 s. Not in our model list Greptile: Grounded F1 33.4%, duration 289 s. Not in our model list Claude Opus 5.5: Grounded F1 73.3%, duration 182 s Claude Opus 5.5: Grounded F1 69.3%, duration 55.1 s OhMyBug: Grounded F1 83%, duration 322 s. Not in our model list intent-review-agent: Grounded F1 82.2%, duration 213 s. Not in our model list OhMyBugintent-review-agentClaude Opus 5.5Claude Opus 5.5
ModelSettingGrounded F1Duration
OhMyBug OhMyBug 83% 322.2 s
intent-review-agent intent-review-agent 82.2% 212.6 s
Copilot code review high, Copilot code review 75.7% 973.5 s
Claude Opus 5.5 Laude RevEngine 73.3% 181.6 s
Copilot code review medium, Copilot code review 72.7% 371.2 s
GPT-5.6 Sol max, Codex 70.2% 567.7 s
GPT-5.6 Sol ultra, Codex 69.9% 543.7 s
Claude Opus 5.5 Klaussy 69.3% 55.1 s
Qodo Qodo 68.3% 274 s
block/review panel=sol-opus · gate=off, block/review 68.2% 196.8 s
HeyDeer depth=deep · filter=detailed, HeyDeer 67.7% 241.9 s
Copilot code review low, Copilot code review 66.2% 275.4 s
Devin AI Devin AI 66.1% 192.9 s
diffox ultra diffox ultra 65.6% 86.8 s
deepseek-flash high, DeepSeek Harness Reviewer 64.8% 319.4 s
DeepSeek-V4-Flash opswiki 64% 318 s
Claude Sonnet 5.5 Klaussy 63.1% 185 s
Cursor Cursor 60.9% 256.5 s
Claude Sonnet 5.5 default, Klaussy 60.2% 176 s
diffox diffox 59.1% 107.5 s
Cubic Cubic 35.8% 701.7 s
Greptile Greptile 33.4% 289.2 s

Number one over time

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

Anthropic
Own leaderboard 1 leader
Claude Opus 5.5 73.3%

Source: Official ReviewBench 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 Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 5.5 GPT-5.6 Sol DeepSeek-V4-Flash 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 5.5 DeepSeek-V4-Flash

Source: Official ReviewBench leaderboard

No lab model card reports it yet

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

More in Coding

All 282 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