Benchmark Almanac

Ramp SWE-Bench v1

Claude Fable 5 88.5% leads since June 2026
Not in model cards
Version 1

Agents modify Ramp’s backend code to satisfy engineering requests. Hidden tests check whether each patch solves the task.

Made by
Ramp Labs
Released
June 2026
Kind
Fixing real repositories
Level
Expert
Scored
Checked automatically
Answers
Code
Tasks
Kept private
Measures
Capability
Language
English
In and out
Text, Code → Code
Interaction
Agentic
Environment
Terminal
Items from
Real work and use
Run by
A host
In views
Agents
Strongest model tested
Claude Opus 5 #11 on the capability index
Last model added
July 2026
Scores refreshed
November 20th, 2026
26
Models scored on its own leaderboard

2 models share the top score of 88.5% among models released by July 2026

Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

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

  1. 1 Claude Fable 5 xhigh 88.5%
  2. 2 Claude Opus 5 high 88.5%
  3. 3 Kimi K3 high 87.2%
  4. 4 Claude Opus 4.7 xhigh 83.3%
  5. 5 GPT-5.5 high 83.3%
  6. 6 GPT-5.6 Sol high 83.3%
  7. 7 GLM-5.2 high 82.1%
  8. 8 Grok 4.5 82.1%
  9. 9 Claude Opus 4.6 high 80.8%
  10. 10 Kimi K2.7 Code high 80.8%
  11. 11 DeepSeek V4 Flash 0731 high 79.5%
  12. 12 Claude Opus 4.8 xhigh 78.2%
  13. 13 GPT-5.6 Terra high 76.9%
  14. 14 Claude Sonnet 5 medium 75.6%
  15. 15 Gemini 3.1 Pro high 74.4%
  16. 16 GPT-5.4 high 74.4%
  17. 17 GPT-5.6 Luna high 74.4%
  18. 18 Claude Sonnet 4.6 medium 73.1%
  19. 19 Kimi K2.6 high 73.1%
  20. 20 GLM-5.1 high 70.5%
  21. 21 Qwen 3.6 Plus high 66.7%
  22. 22 DeepSeek-V4-Pro high 65.4%
  23. 23 Qwen3.7-Plus high 62.8%
  24. 24 GPT-5.4 Mini high 60.3%
  25. 25 Claude Haiku 4.5 high 50%
  26. 26 GPT-5.4 Nano medium 50%

Source: Official Ramp SWE-Bench v1 leaderboard

#ModelScoreSettingReleasedRun onSource
1 Claude Fable 5 88.5% xhigh, mini-SWE-agent Jun 2026 Jun 2026 Official leaderboard
2 Claude Opus 5 88.5% high, mini-SWE-agent Jul 2026 Jul 2026 Official leaderboard
3 Kimi K3 87.2% high, mini-SWE-agent Jul 2026 Jul 2026 Official leaderboard
4 Claude Opus 4.7 83.3% xhigh, mini-SWE-agent Apr 2026 Jun 2026 Official leaderboard
5 GPT-5.5 83.3% high, mini-SWE-agent Apr 2026 Jun 2026 Official leaderboard
6 GPT-5.6 Sol 83.3% high, mini-SWE-agent Jul 2026 Jun 2026 Official leaderboard
7 GLM-5.2 82.1% high, mini-SWE-agent Jun 2026 Jun 2026 Official leaderboard
8 Grok 4.5 82.1% mini-SWE-agent Jul 2026 Jul 2026 Official leaderboard
9 Claude Opus 4.6 80.8% high, mini-SWE-agent Feb 2026 Jun 2026 Official leaderboard
10 Kimi K2.7 Code 80.8% high, mini-SWE-agent Jun 2026 Jun 2026 Official leaderboard
11 DeepSeek V4 Flash 0731 79.5% high, mini-SWE-agent Jul 2026 Aug 2026 Official leaderboard
12 Claude Opus 4.8 78.2% xhigh, mini-SWE-agent May 2026 Jun 2026 Official leaderboard
13 GPT-5.6 Terra 76.9% high, mini-SWE-agent Jul 2026 Jun 2026 Official leaderboard
14 Claude Sonnet 5 75.6% medium, mini-SWE-agent Jun 2026 Jun 2026 Official leaderboard
15 Gemini 3.1 Pro 74.4% high, mini-SWE-agent Feb 2026 Jun 2026 Official leaderboard
16 GPT-5.4 74.4% high, mini-SWE-agent Mar 2026 Jun 2026 Official leaderboard
17 GPT-5.6 Luna 74.4% high, mini-SWE-agent Jul 2026 Jun 2026 Official leaderboard
18 Claude Sonnet 4.6 73.1% medium, mini-SWE-agent Feb 2026 Jun 2026 Official leaderboard
19 Kimi K2.6 73.1% high, mini-SWE-agent Apr 2026 Jun 2026 Official leaderboard
20 GLM-5.1 70.5% high, mini-SWE-agent Apr 2026 Jun 2026 Official leaderboard
21 Qwen 3.6 Plus 66.7% high, mini-SWE-agent Mar 2026 Jun 2026 Official leaderboard
22 DeepSeek-V4-Pro 65.4% high, mini-SWE-agent Apr 2026 Jun 2026 Official leaderboard
23 Qwen3.7-Plus 62.8% high, mini-SWE-agent Jun 2026 Jun 2026 Official leaderboard
24 GPT-5.4 Mini 60.3% high, mini-SWE-agent Mar 2026 Jun 2026 Official leaderboard
25 Claude Haiku 4.5 50% high, mini-SWE-agent Oct 2025 Jun 2026 Official leaderboard
26 GPT-5.4 Nano 50% medium, mini-SWE-agent Mar 2026 Jun 2026 Official leaderboard

Score versus spend

The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

60% 80% 100% $0.01$0.1$1$10 Mean cost per task ($), log scale, lower is better Resolved Claude Opus 5 (high)Kimi K3 (high)GPT-5.6 Sol (high)Kimi K2.7 Code (high)DeepSeek V4 Flash 0731 (high)GPT-5.6 Luna (high) 60% 80% 100% $0.01$0.1$1$10 Mean cost per task ($), log scale, lower is better Resolved Claude Opus 5 (high)Kimi K3 (high)GPT-5.6 Sol (high)Kimi K2.7 Code (high)DeepSeek V4 Flash 0731 (high)GPT-5.6 Luna (high)
ModelSettingResolvedMean cost per task
Claude Opus 5 high, mini-SWE-agent 88.5% $1.84
Claude Fable 5 xhigh, mini-SWE-agent 88.5% $2.62
Kimi K3 high, mini-SWE-agent 87.2% $1.6
GPT-5.6 Sol high, mini-SWE-agent 83.3% $0.99
GPT-5.5 high, mini-SWE-agent 83.3% $1.83
Claude Opus 4.7 xhigh, mini-SWE-agent 83.3% $2.26
Grok 4.5 mini-SWE-agent 82.1% $1.09
GLM-5.2 high, mini-SWE-agent 82.1% $1.84
Kimi K2.7 Code high, mini-SWE-agent 80.8% $0.88
Claude Opus 4.6 high, mini-SWE-agent 80.8% $1.41
DeepSeek V4 Flash 0731 high, mini-SWE-agent 79.5% $0.12
Claude Opus 4.8 xhigh, mini-SWE-agent 78.2% $1.09
GPT-5.6 Terra high, mini-SWE-agent 76.9% $0.26
Claude Sonnet 5 medium, mini-SWE-agent 75.6% $1.19
GPT-5.6 Luna high, mini-SWE-agent 74.4% $0.04
GPT-5.4 high, mini-SWE-agent 74.4% $0.64
Gemini 3.1 Pro high, mini-SWE-agent 74.4% $1.03
Kimi K2.6 high, mini-SWE-agent 73.1% $0.69
Claude Sonnet 4.6 medium, mini-SWE-agent 73.1% $0.72
GLM-5.1 high, mini-SWE-agent 70.5% $1.08
Qwen 3.6 Plus high, mini-SWE-agent 66.7% $0.29
DeepSeek-V4-Pro high, mini-SWE-agent 65.4% $0.81
Qwen3.7-Plus high, mini-SWE-agent 62.8% $0.15
GPT-5.4 Mini high, mini-SWE-agent 60.3% $0.22
GPT-5.4 Nano medium, mini-SWE-agent 50% $0.09
Claude Haiku 4.5 high, mini-SWE-agent 50% $0.49

The best score, 88.5%, costs $1.84 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Moonshot AI OpenAI Anthropic DeepSeek Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale Claude Opus 5, high GPT-5.6 Sol, high DeepSeek V4 Flash 0731, high GPT-5.6 Luna, high 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale Claude Opus 5, high Kimi K2.7 Code, high

Source: Official Ramp SWE-Bench v1 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Fable 5 xhigh mini-SWE-agent 88.5% $2.62 15,196
Claude Opus 5 high mini-SWE-agent 88.5% $1.84 17,758
Kimi K3 high mini-SWE-agent 87.2% $1.60 26,516
Claude Opus 4.7 xhigh mini-SWE-agent 83.3% $2.26 17,999
GPT-5.5 high mini-SWE-agent 83.3% $1.83 19,825
GPT-5.6 Sol high mini-SWE-agent 83.3% $0.99 8,908
GLM-5.2 high mini-SWE-agent 82.1% $1.84 34,001
Grok 4.5 mini-SWE-agent 82.1% $1.09 24,653
Claude Opus 4.6 high mini-SWE-agent 80.8% $1.41 10,495
Kimi K2.7 Code high mini-SWE-agent 80.8% $0.88 19,640
DeepSeek V4 Flash 0731 high mini-SWE-agent 79.5% $0.12 27,167
Claude Opus 4.8 xhigh mini-SWE-agent 78.2% $1.09 13,052
GPT-5.6 Terra high mini-SWE-agent 76.9% $0.26 6,340
Claude Sonnet 5 medium mini-SWE-agent 75.6% $1.19 15,376
Gemini 3.1 Pro high mini-SWE-agent 74.4% $1.03 11,303
GPT-5.4 high mini-SWE-agent 74.4% $0.64 13,439
GPT-5.6 Luna high mini-SWE-agent 74.4% $0.04 8,564
Claude Sonnet 4.6 medium mini-SWE-agent 73.1% $0.72 11,975
Kimi K2.6 high mini-SWE-agent 73.1% $0.69 19,316
GLM-5.1 high mini-SWE-agent 70.5% $1.08 11,568
Qwen 3.6 Plus high mini-SWE-agent 66.7% $0.29 26,519
DeepSeek-V4-Pro high mini-SWE-agent 65.4% $0.81 12,880
Qwen3.7-Plus high mini-SWE-agent 62.8% $0.15 19,157
GPT-5.4 Mini high mini-SWE-agent 60.3% $0.22 24,204
Claude Haiku 4.5 high mini-SWE-agent 50% $0.49 14,857
GPT-5.4 Nano medium mini-SWE-agent 50% $0.09 27,233

2 models have held number one on the official leaderboard since its release in June 2026

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

Anthropic
Own leaderboard Passed 85% in Jun 2026, by Claude Fable 5

Source: Official Ramp SWE-Bench v1 leaderboard

Claude Fable 5 on the official leaderboard passed 85% in June 2026

Models released before it came out already scored that high. All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Fable 5 Claude Opus 4.7 Claude Haiku 4.5 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Fable 5 Claude Opus 4.6

Source: Official Ramp SWE-Bench v1 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.

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

Anthropic Google DeepMind OpenAI
Ramp SWE-Bench v2 4 leaders
Ramp SWE-Bench v1 Passed 85% in Jun 2026, by Claude Fable 5

Source: Official leaderboards

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

All 9 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access The tasks are private.
  • Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
  • Pass Independent runs Scores from the official leaderboard (26 entries).
  • Pass Models scored 26 models we track.
  • Pass Newest score The newest is from July 2026, for DeepSeek V4 Flash 0731.
  • Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is Claude Opus 5 (#11).
  • Note Headroom Saturated. The best is 88.5%, past 85% since June 2026.
Leaderboard import 3 pass
  • Pass Importer We read the board on labs.ramp.com: 26 rows for models we track 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