Benchmark Almanac

Ramp SWE-Bench v2

GPT-6.2 Astra 76.4% leads since November 2026 GPT-6.2 Astra 75% leads in mini-SWE-agent
Not in model cards
Version 2

Agents solve 500 engineering tasks from Ramp's production backend. A panel of model judges compares each patch with the code engineers merged.

Ramp runs every model in a shared Bash harness (mini-SWE-agent, v2.0) and the labs' own coding agents in their native harnesses (v2.1) on the same tasks.

Made by
Ramp Labs
Released
November 2026
Kind
Fixing real repositories
Level
Expert
Scored
Graded by a model
Answers
Code
Tasks
Kept private
Measures
Capability
Language
English
In and out
Text, Code → Code, Actions
Interaction
Agentic
Environment
Terminal
Items from
Real work and use
Run by
A host
In views
Agents
Strongest model tested
GPT-6.2 Astra #1 on the capability index
Last model added
November 2026
Scores refreshed
November 20th, 2026
25
Entries on its own leaderboard, 22 models we track

GPT-6.2 Astra leads with 76.4% GPT-6.2 Astra leads with 76.4% GPT-6.2 Astra leads with 75% in mini-SWE-agent

Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran on its own with several settings, we keep its best. Every run is in the next chart. Agents other companies built around a model get a pale shade of that model's lab color. Gray marks blends and entries not in our model list, such as fine-tunes and older models. None of these count toward leaders. Agents built around a model, such as OpenCode + Gemini 4 Carbon, are their own entries. We don't count them as the model's score. This view shows only models we track, each at its best. This view keeps only runs in mini-SWE-agent, a harness no model’s maker built. A maker’s own agent, such as Claude Code with a Claude model, is left out.

  1. 1 GPT-6.2 Astra high 76.4%
  2. 2 OpenCode + Gemini 4 Carbon high 75.4%
  3. 3 Gemini 4 Carbon high 74%
  4. 4 Claude Fable 5.2 high 73.4%
  5. 5 GPT-6.2 Sol high 73.4%
  6. 6 OpenCode + Gemini 4 Hydrogen high 71.6%
  7. 7 Claude Opus 5.5 high 71.2%
  8. 8 Gemini 4 Hydrogen high 70.8%
  9. 9 GPT-6.1 Terra high 70.4%
  10. 10 Grok 4.7 high 64.4%
  11. 11 Claude Fable 5.1 high 63.8%
  12. 12 GPT-6 Astra high 63.4%
  13. 13 Claude Sonnet 5.5 high 63.2%
  14. 14 DeepSeek Harness + DeepSeek V4.1 Flash 1017 high 60.8%
  15. 15 GPT-6.1 Sol high 59.6%
  16. 16 Muse Wonder high 58.6%
  17. 17 Claude Haiku 5.5 high 57.4%
  18. 18 Gemini 4 Argon high 54.8%
  19. 19 Muse Ember high 54.4%
  20. 20 DeepSeek V4.1 Flash 1017 high 53.4%
  21. 21 GPT-6 Luna high 52.4%
  22. 22 Gemini 3.8 Flash high 47%
  23. 23 Kimi K3.1 Pro high 45.4%
  24. 24 Beam 1.0 high 34%
  25. 25 Nemotron 3.5 Super VL high 12%

Source: Official Ramp SWE-Bench v2 leaderboard

#EntryScore ±SettingReleasedRun onSource
1 GPT-6.2 Astra 76.4% 72.5 to 79.9 high, Codex Nov 2026 Nov 2026 Official leaderboard
2 OpenCode + Gemini 4 Carbon 75.4% 71.4 to 79 high Oct 2026 Nov 2026 Official leaderboard
3 Gemini 4 Carbon 74% 70 to 77.7 high, mini-SWE-agent Oct 2026 Nov 2026 Official leaderboard
4 Claude Fable 5.2 73.4% 69.4 to 77.1 high, Claude Code Nov 2026 Nov 2026 Official leaderboard
5 GPT-6.2 Sol 73.4% 69.4 to 77.1 high, Codex Nov 2026 Nov 2026 Official leaderboard
6 OpenCode + Gemini 4 Hydrogen 71.6% 67.5 to 75.4 high Nov 2026 Nov 2026 Official leaderboard
7 Claude Opus 5.5 71.2% 67.1 to 75 high, mini-SWE-agent Sep 2026 Nov 2026 Official leaderboard
8 Gemini 4 Hydrogen 70.8% 66.7 to 74.6 high, mini-SWE-agent Nov 2026 Nov 2026 Official leaderboard
9 GPT-6.1 Terra 70.4% 66.3 to 74.2 high, Codex Oct 2026 Nov 2026 Official leaderboard
10 Grok 4.7 64.4% 60.1 to 68.5 high, Grok Build Sep 2026 Nov 2026 Official leaderboard
11 Claude Fable 5.1 63.8% 59.5 to 67.9 high, mini-SWE-agent Sep 2026 Nov 2026 Official leaderboard
12 GPT-6 Astra 63.4% 59.1 to 67.5 high, Codex Sep 2026 Nov 2026 Official leaderboard
13 Claude Sonnet 5.5 63.2% 58.9 to 67.3 high, Claude Code Sep 2026 Nov 2026 Official leaderboard
14 DeepSeek Harness + DeepSeek V4.1 Flash 1017 60.8% ±4.3 high Oct 2026 Nov 2026 Official leaderboard
15 GPT-6.1 Sol 59.6% 55.2 to 63.8 high, Codex Sep 2026 Nov 2026 Official leaderboard
16 Muse Wonder 58.6% 54.2 to 62.8 high, Muse Code Oct 2026 Nov 2026 Official leaderboard
17 Claude Haiku 5.5 57.4% ±4.4 high, Claude Code Oct 2026 Nov 2026 Official leaderboard
18 Gemini 4 Argon 54.8% ±4.4 high, mini-SWE-agent Sep 2026 Nov 2026 Official leaderboard
19 Muse Ember 54.4% ±4.4 high, mini-SWE-agent Nov 2026 Nov 2026 Official leaderboard
20 DeepSeek V4.1 Flash 1017 53.4% ±4.4 high, mini-SWE-agent Oct 2026 Nov 2026 Official leaderboard
21 GPT-6 Luna 52.4% ±4.4 high, Codex Sep 2026 Nov 2026 Official leaderboard
22 Gemini 3.8 Flash 47% ±4.3 high, mini-SWE-agent Sep 2026 Nov 2026 Official leaderboard
23 Kimi K3.1 Pro 45.4% ±4.3 high, mini-SWE-agent Oct 2026 Nov 2026 Official leaderboard
24 Beam 1.0 34% 30 to 38.3 high, mini-SWE-agent Oct 2026 Nov 2026 Official leaderboard
25 Nemotron 3.5 Super VL 12% 9.4 to 15.1 high, mini-SWE-agent Oct 2026 Nov 2026 Official leaderboard
#ModelScore ±SettingReleasedRun onSource
1 GPT-6.2 Astra 76.4% 72.5 to 79.9 high, Codex Nov 2026 Nov 2026 Official leaderboard
2 Gemini 4 Carbon 74% 70 to 77.7 high, mini-SWE-agent Oct 2026 Nov 2026 Official leaderboard
3 Claude Fable 5.2 73.4% 69.4 to 77.1 high, Claude Code Nov 2026 Nov 2026 Official leaderboard
4 GPT-6.2 Sol 73.4% 69.4 to 77.1 high, Codex Nov 2026 Nov 2026 Official leaderboard
5 Claude Opus 5.5 71.2% 67.1 to 75 high, mini-SWE-agent Sep 2026 Nov 2026 Official leaderboard
6 Gemini 4 Hydrogen 70.8% 66.7 to 74.6 high, mini-SWE-agent Nov 2026 Nov 2026 Official leaderboard
7 GPT-6.1 Terra 70.4% 66.3 to 74.2 high, Codex Oct 2026 Nov 2026 Official leaderboard
8 Grok 4.7 64.4% 60.1 to 68.5 high, Grok Build Sep 2026 Nov 2026 Official leaderboard
9 Claude Fable 5.1 63.8% 59.5 to 67.9 high, mini-SWE-agent Sep 2026 Nov 2026 Official leaderboard
10 GPT-6 Astra 63.4% 59.1 to 67.5 high, Codex Sep 2026 Nov 2026 Official leaderboard
11 Claude Sonnet 5.5 63.2% 58.9 to 67.3 high, Claude Code Sep 2026 Nov 2026 Official leaderboard
12 GPT-6.1 Sol 59.6% 55.2 to 63.8 high, Codex Sep 2026 Nov 2026 Official leaderboard
13 Muse Wonder 58.6% 54.2 to 62.8 high, Muse Code Oct 2026 Nov 2026 Official leaderboard
14 Claude Haiku 5.5 57.4% ±4.4 high, Claude Code Oct 2026 Nov 2026 Official leaderboard
15 Gemini 4 Argon 54.8% ±4.4 high, mini-SWE-agent Sep 2026 Nov 2026 Official leaderboard
16 Muse Ember 54.4% ±4.4 high, mini-SWE-agent Nov 2026 Nov 2026 Official leaderboard
17 DeepSeek V4.1 Flash 1017 53.4% ±4.4 high, mini-SWE-agent Oct 2026 Nov 2026 Official leaderboard
18 GPT-6 Luna 52.4% ±4.4 high, Codex Sep 2026 Nov 2026 Official leaderboard
19 Gemini 3.8 Flash 47% ±4.3 high, mini-SWE-agent Sep 2026 Nov 2026 Official leaderboard
20 Kimi K3.1 Pro 45.4% ±4.3 high, mini-SWE-agent Oct 2026 Nov 2026 Official leaderboard
21 Beam 1.0 34% 30 to 38.3 high, mini-SWE-agent Oct 2026 Nov 2026 Official leaderboard
22 Nemotron 3.5 Super VL 12% 9.4 to 15.1 high, mini-SWE-agent Oct 2026 Nov 2026 Official leaderboard
#ModelScore ±SettingReleasedRun onSource
1 GPT-6.2 Astra 75% 71 to 78.6 high, mini-SWE-agent Nov 2026 Nov 2026 Official leaderboard
2 Gemini 4 Carbon 74% 70 to 77.7 high, mini-SWE-agent Oct 2026 Nov 2026 Official leaderboard
3 Claude Fable 5.2 73% 68.9 to 76.7 high, mini-SWE-agent Nov 2026 Nov 2026 Official leaderboard
4 GPT-6.2 Sol 72.4% 68.3 to 76.1 high, mini-SWE-agent Nov 2026 Nov 2026 Official leaderboard
5 Claude Opus 5.5 71.2% 67.1 to 75 high, mini-SWE-agent Sep 2026 Nov 2026 Official leaderboard
6 Gemini 4 Hydrogen 70.8% 66.7 to 74.6 high, mini-SWE-agent Nov 2026 Nov 2026 Official leaderboard
7 Claude Fable 5.1 63.8% 59.5 to 67.9 high, mini-SWE-agent Sep 2026 Nov 2026 Official leaderboard
8 GPT-6.1 Terra 60.6% ±4.3 high, mini-SWE-agent Oct 2026 Nov 2026 Official leaderboard
9 Muse Wonder 57.8% ±4.4 high, mini-SWE-agent Oct 2026 Nov 2026 Official leaderboard
10 Claude Sonnet 5.5 57.6% ±4.3 high, mini-SWE-agent Sep 2026 Nov 2026 Official leaderboard
11 Grok 4.7 55.4% ±4.4 high, mini-SWE-agent Sep 2026 Nov 2026 Official leaderboard
12 Gemini 4 Argon 54.8% ±4.4 high, mini-SWE-agent Sep 2026 Nov 2026 Official leaderboard
13 GPT-6 Astra 54.6% ±4.3 high, mini-SWE-agent Sep 2026 Nov 2026 Official leaderboard
14 Muse Ember 54.4% ±4.4 high, mini-SWE-agent Nov 2026 Nov 2026 Official leaderboard
15 DeepSeek V4.1 Flash 1017 53.4% ±4.4 high, mini-SWE-agent Oct 2026 Nov 2026 Official leaderboard
16 Claude Haiku 5.5 52.8% ±4.4 high, mini-SWE-agent Oct 2026 Nov 2026 Official leaderboard
17 GPT-6.1 Sol 52% ±4.3 high, mini-SWE-agent Sep 2026 Nov 2026 Official leaderboard
18 Gemini 3.8 Flash 47% ±4.3 high, mini-SWE-agent Sep 2026 Nov 2026 Official leaderboard
19 Kimi K3.1 Pro 45.4% ±4.3 high, mini-SWE-agent Oct 2026 Nov 2026 Official leaderboard
20 GPT-6 Luna 42.6% ±4.4 high, mini-SWE-agent Sep 2026 Nov 2026 Official leaderboard
21 Beam 1.0 34% 30 to 38.3 high, mini-SWE-agent Oct 2026 Nov 2026 Official leaderboard
22 Nemotron 3.5 Super VL 12% 9.4 to 15.1 high, mini-SWE-agent Oct 2026 Nov 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.

0% 20% 40% 60% 80% $0.01$0.1$1$10 Mean cost per task ($), log scale, lower is better Resolved GPT-6.2 Astra (high)GPT-6.2 Astra (high)GPT-6.2 Sol (high)DeepSeek Harness + DeepSeek V4.1 Flash 1017 (high)GPT-6.1 Terra (high)Muse Ember (high)GPT-6 Luna (high) 0% 20% 40% 60% 80% $0.01$0.1$1$10 Mean cost per task ($), log scale, lower is better Resolved GPT-6.2 Astra (high)GPT-6.2 Astra (high)GPT-6.2 Sol (high)DeepSeek Harness + DeepSeek V4.1 Flash 1017 (high)GPT-6.1 Terra (high)Muse Ember (high)GPT-6 Luna (high)
ModelSettingResolvedMean cost per task
GPT-6.2 Astra high, Codex 76.4% $3.31
OpenCode + Gemini 4 Carbon high, OpenCode 75.4% $4.25
GPT-6.2 Astra high, mini-SWE-agent 75% $1.03
Gemini 4 Carbon high, mini-SWE-agent 74% $5.33
GPT-6.2 Sol high, Codex 73.4% $1.05
Claude Fable 5.2 high, Claude Code 73.4% $3.79
Claude Fable 5.2 high, mini-SWE-agent 73% $6.45
GPT-6.2 Sol high, mini-SWE-agent 72.4% $0.28
OpenCode + Gemini 4 Hydrogen high, OpenCode 71.6% $2.11
Claude Opus 5.5 high, mini-SWE-agent 71.2% $0.99
Gemini 4 Hydrogen high, mini-SWE-agent 70.8% $1.75
GPT-6.1 Terra high, Codex 70.4% $0.46
Claude Opus 5.5 high, Claude Code 67.8% $1.53
Grok 4.7 high, Grok Build 64.4% $4.32
Claude Fable 5.1 high, mini-SWE-agent 63.8% $2.43
GPT-6 Astra high, Codex 63.4% $5.22
Claude Sonnet 5.5 high, Claude Code 63.2% $0.47
DeepSeek Harness + DeepSeek V4.1 Flash 1017 high, DeepSeek Harness 60.8% $0.22
GPT-6.1 Terra high, mini-SWE-agent 60.6% $0.13
GPT-6.1 Sol high, Codex 59.6% $0.74
Muse Wonder high, Muse Code 58.6% $4.59
Muse Wonder high, mini-SWE-agent 57.8% $2.65
Claude Sonnet 5.5 high, mini-SWE-agent 57.6% $0.24
Claude Haiku 5.5 high, Claude Code 57.4% $0.28
Grok 4.7 high, mini-SWE-agent 55.4% $3.98
Gemini 4 Argon high, mini-SWE-agent 54.8% $6.51
GPT-6 Astra high, mini-SWE-agent 54.6% $1.69
Muse Ember high, mini-SWE-agent 54.4% $0.06
DeepSeek V4.1 Flash 1017 high, mini-SWE-agent 53.4% $0.16
Claude Haiku 5.5 high, mini-SWE-agent 52.8% $0.12
GPT-6 Luna high, Codex 52.4% $0.07
GPT-6.1 Sol high, mini-SWE-agent 52% $0.27
Muse Ember high, Muse Code 51.8% $0.11
Gemini 3.8 Flash high, mini-SWE-agent 47% $2.62
Kimi K3.1 Pro high, mini-SWE-agent 45.4% $0.7
GPT-6 Luna high, mini-SWE-agent 42.6% $0.02
Beam 1.0 high, mini-SWE-agent 34% $0.08
Nemotron 3.5 Super VL high, mini-SWE-agent 12% $0.11

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.

0% 20% 40% 60% 80% $0.01$0.1$1$10 Mean cost per task ($), log scale, lower is better Resolved GPT-6.2 Astra (high)GPT-6.2 Astra (high)GPT-6.2 Sol (high)GPT-6.1 Terra (high)Muse Ember (high)GPT-6 Luna (high) 0% 20% 40% 60% 80% $0.01$0.1$1$10 Mean cost per task ($), log scale, lower is better Resolved GPT-6.2 Astra (high)GPT-6.2 Astra (high)GPT-6.2 Sol (high)GPT-6.1 Terra (high)Muse Ember (high)GPT-6 Luna (high)
ModelSettingResolvedMean cost per task
GPT-6.2 Astra high, Codex 76.4% $3.31
GPT-6.2 Astra high, mini-SWE-agent 75% $1.03
Gemini 4 Carbon high, mini-SWE-agent 74% $5.33
GPT-6.2 Sol high, Codex 73.4% $1.05
Claude Fable 5.2 high, Claude Code 73.4% $3.79
Claude Fable 5.2 high, mini-SWE-agent 73% $6.45
GPT-6.2 Sol high, mini-SWE-agent 72.4% $0.28
Claude Opus 5.5 high, mini-SWE-agent 71.2% $0.99
Gemini 4 Hydrogen high, mini-SWE-agent 70.8% $1.75
GPT-6.1 Terra high, Codex 70.4% $0.46
Claude Opus 5.5 high, Claude Code 67.8% $1.53
Grok 4.7 high, Grok Build 64.4% $4.32
Claude Fable 5.1 high, mini-SWE-agent 63.8% $2.43
GPT-6 Astra high, Codex 63.4% $5.22
Claude Sonnet 5.5 high, Claude Code 63.2% $0.47
GPT-6.1 Terra high, mini-SWE-agent 60.6% $0.13
GPT-6.1 Sol high, Codex 59.6% $0.74
Muse Wonder high, Muse Code 58.6% $4.59
Muse Wonder high, mini-SWE-agent 57.8% $2.65
Claude Sonnet 5.5 high, mini-SWE-agent 57.6% $0.24
Claude Haiku 5.5 high, Claude Code 57.4% $0.28
Grok 4.7 high, mini-SWE-agent 55.4% $3.98
Gemini 4 Argon high, mini-SWE-agent 54.8% $6.51
GPT-6 Astra high, mini-SWE-agent 54.6% $1.69
Muse Ember high, mini-SWE-agent 54.4% $0.06
DeepSeek V4.1 Flash 1017 high, mini-SWE-agent 53.4% $0.16
Claude Haiku 5.5 high, mini-SWE-agent 52.8% $0.12
GPT-6 Luna high, Codex 52.4% $0.07
GPT-6.1 Sol high, mini-SWE-agent 52% $0.27
Muse Ember high, Muse Code 51.8% $0.11
Gemini 3.8 Flash high, mini-SWE-agent 47% $2.62
Kimi K3.1 Pro high, mini-SWE-agent 45.4% $0.7
GPT-6 Luna high, mini-SWE-agent 42.6% $0.02
Beam 1.0 high, mini-SWE-agent 34% $0.08
Nemotron 3.5 Super VL high, mini-SWE-agent 12% $0.11

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.

0% 20% 40% 60% 80% $0.01$0.1$1$10 Mean cost per task ($), log scale, lower is better Resolved GPT-6.2 Astra (high)GPT-6.2 Sol (high)GPT-6.1 Terra (high)Muse Ember (high)GPT-6 Luna (high) 0% 20% 40% 60% 80% $0.01$0.1$1$10 Mean cost per task ($), log scale, lower is better Resolved GPT-6.2 Astra (high)GPT-6.2 Sol (high)GPT-6.1 Terra (high)Muse Ember (high)GPT-6 Luna (high)
ModelSettingResolvedMean cost per task
GPT-6.2 Astra high, mini-SWE-agent 75% $1.03
Gemini 4 Carbon high, mini-SWE-agent 74% $5.33
Claude Fable 5.2 high, mini-SWE-agent 73% $6.45
GPT-6.2 Sol high, mini-SWE-agent 72.4% $0.28
Claude Opus 5.5 high, mini-SWE-agent 71.2% $0.99
Gemini 4 Hydrogen high, mini-SWE-agent 70.8% $1.75
Claude Fable 5.1 high, mini-SWE-agent 63.8% $2.43
GPT-6.1 Terra high, mini-SWE-agent 60.6% $0.13
Muse Wonder high, mini-SWE-agent 57.8% $2.65
Claude Sonnet 5.5 high, mini-SWE-agent 57.6% $0.24
Grok 4.7 high, mini-SWE-agent 55.4% $3.98
Gemini 4 Argon high, mini-SWE-agent 54.8% $6.51
GPT-6 Astra high, mini-SWE-agent 54.6% $1.69
Muse Ember high, mini-SWE-agent 54.4% $0.06
DeepSeek V4.1 Flash 1017 high, mini-SWE-agent 53.4% $0.16
Claude Haiku 5.5 high, mini-SWE-agent 52.8% $0.12
GPT-6.1 Sol high, mini-SWE-agent 52% $0.27
Gemini 3.8 Flash high, mini-SWE-agent 47% $2.62
Kimi K3.1 Pro high, mini-SWE-agent 45.4% $0.7
GPT-6 Luna high, mini-SWE-agent 42.6% $0.02
Beam 1.0 high, mini-SWE-agent 34% $0.08
Nemotron 3.5 Super VL high, mini-SWE-agent 12% $0.11

The best score, 76.4%, costs $3.31 a task

Runs on the cost frontier score higher than any run that's cheaper. GPT-6.2 Sol at high comes within 4 points of the best for $0.28 a task, 12 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Meta Agents other companies built, in a pale shade of the model’s lab 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 GPT-6.2 Astra, high GPT-6.2 Astra, high GPT-6.2 Sol, high DeepSeek Harness + DeepSeek V4.1 Flash 1017, high Muse Ember, high GPT-6 Luna, high 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale GPT-6.2 Astra, high DeepSeek Harness + DeepSeek V4.1 Flash 1017, high

Source: Official Ramp SWE-Bench v2 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6.2 Astra high Codex 76.4% $3.31 9,593
OpenCode + Gemini 4 Carbon high OpenCode 75.4% $4.25 72,348
GPT-6.2 Astra high mini-SWE-agent 75% $1.03 7,097
Gemini 4 Carbon high mini-SWE-agent 74% $5.33 73,704
Claude Fable 5.2 high Claude Code 73.4% $3.79 30,618
GPT-6.2 Sol high Codex 73.4% $1.05 10,919
Claude Fable 5.2 high mini-SWE-agent 73% $6.45 63,016
GPT-6.2 Sol high mini-SWE-agent 72.4% $0.28 8,066
OpenCode + Gemini 4 Hydrogen high OpenCode 71.6% $2.11 48,264
Claude Opus 5.5 high mini-SWE-agent 71.2% $0.99 13,707
Gemini 4 Hydrogen high mini-SWE-agent 70.8% $1.75 47,277
GPT-6.1 Terra high Codex 70.4% $0.46 16,043
Claude Opus 5.5 high Claude Code 67.8% $1.53 16,447
Grok 4.7 high Grok Build 64.4% $4.32 50,767
Claude Fable 5.1 high mini-SWE-agent 63.8% $2.43 18,100
GPT-6 Astra high Codex 63.4% $5.22 14,662
Claude Sonnet 5.5 high Claude Code 63.2% $0.47 10,602
DeepSeek Harness + DeepSeek V4.1 Flash 1017 high DeepSeek Harness 60.8% $0.22 40,170
GPT-6.1 Terra high mini-SWE-agent 60.6% $0.13 7,661
GPT-6.1 Sol high Codex 59.6% $0.74 16,113
Muse Wonder high Muse Code 58.6% $4.59 38,797
Muse Wonder high mini-SWE-agent 57.8% $2.65 26,331
Claude Sonnet 5.5 high mini-SWE-agent 57.6% $0.24 6,798
Claude Haiku 5.5 high Claude Code 57.4% $0.28 35,854
Grok 4.7 high mini-SWE-agent 55.4% $3.98 17,674
Gemini 4 Argon high mini-SWE-agent 54.8% $6.51 92,379
GPT-6 Astra high mini-SWE-agent 54.6% $1.69 9,030
Muse Ember high mini-SWE-agent 54.4% $0.06 28,963
DeepSeek V4.1 Flash 1017 high mini-SWE-agent 53.4% $0.16 22,857
Claude Haiku 5.5 high mini-SWE-agent 52.8% $0.12 26,909
GPT-6 Luna high Codex 52.4% $0.07 28,442
GPT-6.1 Sol high mini-SWE-agent 52% $0.27 9,111
Muse Ember high Muse Code 51.8% $0.11 41,751
Gemini 3.8 Flash high mini-SWE-agent 47% $2.62 80,934
Kimi K3.1 Pro high mini-SWE-agent 45.4% $0.70 11,519
GPT-6 Luna high mini-SWE-agent 42.6% $0.02 11,048
Beam 1.0 high mini-SWE-agent 34% $0.08 8,639
Nemotron 3.5 Super VL high mini-SWE-agent 12% $0.11 11,766

The best score, 76.4%, costs $3.31 a task

Runs on the cost frontier score higher than any run that's cheaper. GPT-6.2 Sol at high comes within 4 points of the best for $0.28 a task, 12 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Meta 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 GPT-6.2 Astra, high GPT-6.2 Astra, high GPT-6.2 Sol, high GPT-6.1 Terra, high Muse Ember, high GPT-6 Luna, high 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale GPT-6.2 Astra, high GPT-6.1 Terra, high

Source: Official Ramp SWE-Bench v2 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6.2 Astra high Codex 76.4% $3.31 9,593
GPT-6.2 Astra high mini-SWE-agent 75% $1.03 7,097
Gemini 4 Carbon high mini-SWE-agent 74% $5.33 73,704
Claude Fable 5.2 high Claude Code 73.4% $3.79 30,618
GPT-6.2 Sol high Codex 73.4% $1.05 10,919
Claude Fable 5.2 high mini-SWE-agent 73% $6.45 63,016
GPT-6.2 Sol high mini-SWE-agent 72.4% $0.28 8,066
Claude Opus 5.5 high mini-SWE-agent 71.2% $0.99 13,707
Gemini 4 Hydrogen high mini-SWE-agent 70.8% $1.75 47,277
GPT-6.1 Terra high Codex 70.4% $0.46 16,043
Claude Opus 5.5 high Claude Code 67.8% $1.53 16,447
Grok 4.7 high Grok Build 64.4% $4.32 50,767
Claude Fable 5.1 high mini-SWE-agent 63.8% $2.43 18,100
GPT-6 Astra high Codex 63.4% $5.22 14,662
Claude Sonnet 5.5 high Claude Code 63.2% $0.47 10,602
GPT-6.1 Terra high mini-SWE-agent 60.6% $0.13 7,661
GPT-6.1 Sol high Codex 59.6% $0.74 16,113
Muse Wonder high Muse Code 58.6% $4.59 38,797
Muse Wonder high mini-SWE-agent 57.8% $2.65 26,331
Claude Sonnet 5.5 high mini-SWE-agent 57.6% $0.24 6,798
Claude Haiku 5.5 high Claude Code 57.4% $0.28 35,854
Grok 4.7 high mini-SWE-agent 55.4% $3.98 17,674
Gemini 4 Argon high mini-SWE-agent 54.8% $6.51 92,379
GPT-6 Astra high mini-SWE-agent 54.6% $1.69 9,030
Muse Ember high mini-SWE-agent 54.4% $0.06 28,963
DeepSeek V4.1 Flash 1017 high mini-SWE-agent 53.4% $0.16 22,857
Claude Haiku 5.5 high mini-SWE-agent 52.8% $0.12 26,909
GPT-6 Luna high Codex 52.4% $0.07 28,442
GPT-6.1 Sol high mini-SWE-agent 52% $0.27 9,111
Muse Ember high Muse Code 51.8% $0.11 41,751
Gemini 3.8 Flash high mini-SWE-agent 47% $2.62 80,934
Kimi K3.1 Pro high mini-SWE-agent 45.4% $0.70 11,519
GPT-6 Luna high mini-SWE-agent 42.6% $0.02 11,048
Beam 1.0 high mini-SWE-agent 34% $0.08 8,639
Nemotron 3.5 Super VL high mini-SWE-agent 12% $0.11 11,766

The best score, 75%, costs $1.03 a task

Runs on the cost frontier score higher than any run that's cheaper. GPT-6.2 Sol at high comes within 2.6 points of the best for $0.28 a task, 4 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Meta 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 GPT-6.2 Astra, high GPT-6.2 Sol, high GPT-6.1 Terra, high Muse Ember, high GPT-6 Luna, high 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale GPT-6.2 Astra, high GPT-6.1 Terra, high

Source: Official Ramp SWE-Bench v2 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6.2 Astra high mini-SWE-agent 75% $1.03 7,097
Gemini 4 Carbon high mini-SWE-agent 74% $5.33 73,704
Claude Fable 5.2 high mini-SWE-agent 73% $6.45 63,016
GPT-6.2 Sol high mini-SWE-agent 72.4% $0.28 8,066
Claude Opus 5.5 high mini-SWE-agent 71.2% $0.99 13,707
Gemini 4 Hydrogen high mini-SWE-agent 70.8% $1.75 47,277
Claude Fable 5.1 high mini-SWE-agent 63.8% $2.43 18,100
GPT-6.1 Terra high mini-SWE-agent 60.6% $0.13 7,661
Muse Wonder high mini-SWE-agent 57.8% $2.65 26,331
Claude Sonnet 5.5 high mini-SWE-agent 57.6% $0.24 6,798
Grok 4.7 high mini-SWE-agent 55.4% $3.98 17,674
Gemini 4 Argon high mini-SWE-agent 54.8% $6.51 92,379
GPT-6 Astra high mini-SWE-agent 54.6% $1.69 9,030
Muse Ember high mini-SWE-agent 54.4% $0.06 28,963
DeepSeek V4.1 Flash 1017 high mini-SWE-agent 53.4% $0.16 22,857
Claude Haiku 5.5 high mini-SWE-agent 52.8% $0.12 26,909
GPT-6.1 Sol high mini-SWE-agent 52% $0.27 9,111
Gemini 3.8 Flash high mini-SWE-agent 47% $2.62 80,934
Kimi K3.1 Pro high mini-SWE-agent 45.4% $0.70 11,519
GPT-6 Luna high mini-SWE-agent 42.6% $0.02 11,048
Beam 1.0 high mini-SWE-agent 34% $0.08 8,639
Nemotron 3.5 Super VL high mini-SWE-agent 12% $0.11 11,766

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

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

Google DeepMind OpenAI
Own leaderboard 2 leaders

Source: Official Ramp SWE-Bench v2 leaderboard

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
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) GPT-6.2 Astra Claude Opus 5.5 Grok 4.7 0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) GPT-6.2 Astra Grok 4.7

Source: Official Ramp SWE-Bench v2 leaderboard

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) GPT-6.2 Astra Claude Opus 5.5 Claude Fable 5.1 0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) GPT-6.2 Astra Claude Fable 5.1

Source: Official Ramp SWE-Bench v2 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.

One logged change that moves scores

Changes by the maintainers, so a score from before a change may not compare with one from after it. Newest first, each with the maintainers' own source. The list may be incomplete. What we log.

  1. Harness changed November 2026

    Ramp added runs in each lab's own coding agent (v2.1) beside the shared Bash harness (v2.0), on the same 500 tasks. Source

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 11 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access The tasks are private.
  • Pass Confidence Checked against a primary source.
Scores 5 pass
  • Pass Independent runs Scores from the official leaderboard (25 entries).
  • Pass Models scored 22 models we track, plus 3 other entries on its leaderboard.
  • Pass Newest score The newest is from November 2026, for GPT-6.2 Astra.
  • Pass Frontier models 10 of the top 10 have a score, led by GPT-6.2 Astra (#1).
  • Pass Headroom The best is 76.4%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on labs.ramp.com: 38 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