SWE-Bench Pro V2 Hard
The official leaderboard hasn’t scored a model released after Inkling 2 Small (November 2026). Labs report their own numbers for one newer model: Inkling 2 Mini at 77.9%.
Agents resolve the 51 SWE-Bench Pro V2 tasks that at least two of five model families failed. Graded with the V2 locked protocol.
The paper reports Hard scores from 52.9 to 98.0 while the full V2 set reaches 99.4. It is the hard config of the V2 dataset, and three of 54 candidates were removed after expert review.
- Released
- September 2026
- Kind
- Fixing real repositories
- Level
- Research frontier
- Scored
- Checked automatically
- Answers
- Code
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Agentic
- Environment
- Terminal
- Items from
- Another benchmark
- Refresh
- New versions replace it
- Run by
- Anyone
- Based on
- SWE-Bench Pro V2
- In views
- Agents
- Strongest model tested
- GPT-6 Astra #4 on the capability index
- Last in a launch post
- November 2026
- Last model added
- November 2026
- Scores refreshed
- November 20th, 2026
In the authors’ words
SWE-Bench Pro is a widely used benchmark of long-horizon software engineering tasks for coding agents. As agents improve, they reach task issues earlier agents never touched and interact with the evaluation environment in ways the original harness did not anticipate. We present SWE-Bench Pro V2, a revised public split built to keep the benchmark accurate as capabilities advance. Agentic audits, expert adjudication and blind re-implementation repaired or retired 480 of the 731 tasks: 23 contracted engineers spent 404 hours, each working in a language they write professionally, and every revised instruction was verified by a second engineer solving the task from the text alone. A hardened protocol restricts network access to the model endpoint and re-grades every patch in a clean container, so agents cannot retrieve upstream fixes or get credit for changes outside their patch. On the 642 V2 tasks frontier agents resolve up to 99.4% while other systems stay clearly separated, and a 51-task Hard split spreads systems from 52.9% to 98.0%. We release the benchmark and distill lessons for maintaining agent benchmarks over time.
Claude Opus 5 leads with 98% Claude Opus 5 leads with 98% Gemini 4 Argon leads with 92.2% in mini-SWE-agent
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. 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.
Claude Opus 5 leads with 98%, claims included
Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model. Striped bars are claims, and dotted bars are reruns by Epoch AI or another board.
- 98.0 Claude Opus 5
- 92.2 Claude Fable 5.1
- 92.2
Gemini 4 Argon
- 90.2 GPT-6 Astra
- 90.2
Muse Wonder
- 88.2 Claude Sonnet 5
- 88.2 Kimi K3
- 86.3 GPT-5.6 Terra
- 86.3
Inkling 2 Small
- 84.3
MiniMax-M3.1-Flash
- 84.3 GLM-5.3
- 82.4 GPT-5.6 Sol
- 82.3
Muse Spark Open
- 70.6 Claude Sonnet 5.5
- 58.8
Gemini 3.8 Flash
- 56.9
Inkling
- 52.9 Claude Haiku 5.5
- 25.5 Claude Haiku 4.5
- 1 Claude Opus 5 xhigh 98%
- 2 Claude Fable 5.1 high 92.2%
- 3
Gemini 4 Argon high 92.2%
- 4 GPT-6 Astra high 90.2%
- 5
Muse Wonder xhigh 90.2%
- 6 Claude Sonnet 5 xhigh 88.2%
- 7 Kimi K3 max 88.2%
- 8 GPT-5.6 Terra xhigh 86.3%
- 9
Inkling 2 Small 86.3%
- 10
MiniMax-M3.1-Flash xhigh 84.3%
- 11 GLM-5.3 max 84.3%
- 12 GPT-5.6 Sol xhigh 82.4%
- 13
Muse Spark Open xhigh 82.3%
- 14 Claude Sonnet 5.5 xhigh 70.6%
- 15
Gemini 3.8 Flash high 58.8%
- 16
Inkling xhigh 56.9%
- 17 Claude Haiku 5.5 high 52.9%
- 18 Claude Haiku 4.5 xhigh 25.5%
Source: Official SWE-Bench Pro V2 Hard leaderboard
- 1
Gemini 4 Argon high 92.2%
- 2
Muse Wonder xhigh 90.2%
- 3 Kimi K3 max 88.2%
- 4
Inkling 2 Small 86.3%
- 5
MiniMax-M3.1-Flash xhigh 84.3%
- 6 GLM-5.3 max 84.3%
- 7
Muse Spark Open xhigh 82.3%
- 8 Claude Sonnet 5.5 xhigh 70.6%
- 9
Gemini 3.8 Flash high 58.8%
- 10
Inkling xhigh 56.9%
- 11 Claude Haiku 5.5 high 52.9%
Source: Official SWE-Bench Pro V2 Hard leaderboard
- 98.0 Claude Opus 5
- 92.2 Claude Fable 5.1
- 92.2
Gemini 4 Argon
- 90.2 GPT-6 Astra
- 90.2
Muse Wonder
- 89.9 SWE-2 Flash
- 88.2 Claude Sonnet 5
- 88.2 Kimi K3
- 86.3 GPT-5.6 Terra
- 86.3
Inkling 2 Small
- 84.3
MiniMax-M3.1-Flash
- 84.3 GLM-5.3
- 82.4 GPT-5.6 Sol
- 82.3
Muse Spark Open
- 77.9
Inkling 2 Mini
- 77.2 Beam 1.0
- 70.6 Claude Sonnet 5.5
- 58.8
Gemini 3.8 Flash
- 56.9
Inkling
- 52.9 Claude Haiku 5.5
- 25.5 Claude Haiku 4.5
- 1 Claude Opus 5 xhigh 98%
- 2 Claude Fable 5.1 high 92.2%
- 3
Gemini 4 Argon high 92.2%
- 4 GPT-6 Astra high 90.2%
- 5
Muse Wonder xhigh 90.2%
- 6 SWE-2 Flash 89.9%
- 7 Claude Sonnet 5 xhigh 88.2%
- 8 Kimi K3 max 88.2%
- 9 GPT-5.6 Terra xhigh 86.3%
- 10
Inkling 2 Small 86.3%
- 11
MiniMax-M3.1-Flash xhigh 84.3%
- 12 GLM-5.3 max 84.3%
- 13 GPT-5.6 Sol xhigh 82.4%
- 14
Muse Spark Open xhigh 82.3%
- 15
Inkling 2 Mini 77.9%
- 16 Beam 1.0 77.2%
- 17 Claude Sonnet 5.5 xhigh 70.6%
- 18
Gemini 3.8 Flash high 58.8%
- 19
Inkling xhigh 56.9%
- 20 Claude Haiku 5.5 high 52.9%
- 21 Claude Haiku 4.5 xhigh 25.5%
Sources: Official SWE-Bench Pro V2 Hard leaderboard and labs’ model cards
| # | Model | Score ± | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5 | 98% | xhigh, Claude Code | Jul 2026 | Official leaderboard |
| 2 | Claude Fable 5.1 | 92.2% | high, Claude Code | Sep 2026 | Official leaderboard |
| 3 | | 92.2% | high, mini-SWE-agent | Sep 2026 | Official leaderboard |
| 4 | GPT-6 Astra | 90.2% | high, Codex | Sep 2026 | Official leaderboard |
| 5 | | 90.2% | xhigh, mini-SWE-agent | Oct 2026 | Official leaderboard |
| 6 | Claude Sonnet 5 | 88.2% | xhigh, Claude Code | Jun 2026 | Official leaderboard |
| 7 | Kimi K3 | 88.2% | max, mini-SWE-agent | Jul 2026 | Official leaderboard |
| 8 | GPT-5.6 Terra | 86.3% | xhigh, Codex | Jul 2026 | Official leaderboard |
| 9 | | 86.3% | Nov 2026 | Official leaderboard | |
| 10 | | 84.3% | xhigh, mini-SWE-agent | Oct 2026 | Official leaderboard |
| 11 | GLM-5.3 | 84.3% | max, mini-SWE-agent | Aug 2026 | Official leaderboard |
| 12 | GPT-5.6 Sol | 82.4% | xhigh, Codex | Jul 2026 | Official leaderboard |
| 13 | | 82.3% | xhigh, mini-SWE-agent | Oct 2026 | Official leaderboard |
| 14 | Claude Sonnet 5.5 | 70.6% | xhigh, mini-SWE-agent | Sep 2026 | Official leaderboard |
| 15 | | 58.8% | high, mini-SWE-agent | Sep 2026 | Official leaderboard |
| 16 | | 56.9% | xhigh, mini-SWE-agent | Jul 2026 | Official leaderboard |
| 17 | Claude Haiku 5.5 | 52.9% ±13 | high, mini-SWE-agent | Oct 2026 | Official leaderboard |
| 18 | Claude Haiku 4.5 | 25.5% | xhigh, Claude Code | Oct 2025 | Official leaderboard |
| # | Model | Score ± | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 92.2% | high, mini-SWE-agent | Sep 2026 | Official leaderboard |
| 2 | | 90.2% | xhigh, mini-SWE-agent | Oct 2026 | Official leaderboard |
| 3 | Kimi K3 | 88.2% | max, mini-SWE-agent | Jul 2026 | Official leaderboard |
| 4 | | 86.3% | Nov 2026 | Official leaderboard | |
| 5 | | 84.3% | xhigh, mini-SWE-agent | Oct 2026 | Official leaderboard |
| 6 | GLM-5.3 | 84.3% | max, mini-SWE-agent | Aug 2026 | Official leaderboard |
| 7 | | 82.3% | xhigh, mini-SWE-agent | Oct 2026 | Official leaderboard |
| 8 | Claude Sonnet 5.5 | 70.6% | xhigh, mini-SWE-agent | Sep 2026 | Official leaderboard |
| 9 | | 58.8% | high, mini-SWE-agent | Sep 2026 | Official leaderboard |
| 10 | | 56.9% | xhigh, mini-SWE-agent | Jul 2026 | Official leaderboard |
| 11 | Claude Haiku 5.5 | 52.9% ±13 | high, mini-SWE-agent | Oct 2026 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5 | 98% | xhigh, Claude Code | Jul 2026 | Official leaderboard |
| 2 | Claude Fable 5.1 | 92.2% | high, Claude Code | Sep 2026 | Official leaderboard |
| 3 | 92.2% | high, mini-SWE-agent | Sep 2026 | Official leaderboard | |
| 4 | GPT-6 Astra | 90.2% | high, Codex | Sep 2026 | Official leaderboard |
| 5 | 90.2% | xhigh, mini-SWE-agent | Oct 2026 | Official leaderboard | |
| 6 | SWE-2 Flash | 89.9% | Nov 2026 | Lab model card | |
| 7 | Claude Sonnet 5 | 88.2% | xhigh, Claude Code | Jun 2026 | Official leaderboard |
| 8 | Kimi K3 | 88.2% | max, mini-SWE-agent | Jul 2026 | Official leaderboard |
| 9 | GPT-5.6 Terra | 86.3% | xhigh, Codex | Jul 2026 | Official leaderboard |
| 10 | 86.3% | Nov 2026 | Lab model card | ||
| 11 | 84.3% | xhigh, mini-SWE-agent | Oct 2026 | Official leaderboard | |
| 12 | GLM-5.3 | 84.3% | max, mini-SWE-agent | Aug 2026 | Official leaderboard |
| 13 | GPT-5.6 Sol | 82.4% | xhigh, Codex | Jul 2026 | Official leaderboard |
| 14 | 82.3% | xhigh, mini-SWE-agent | Oct 2026 | Official leaderboard | |
| 15 | 77.9% | Nov 2026 | Lab model card | ||
| 16 | Beam 1.0 | 77.2% | Oct 2026 | Lab model card | |
| 17 | Claude Sonnet 5.5 | 70.6% | xhigh, mini-SWE-agent | Sep 2026 | Official leaderboard |
| 18 | 58.8% | high, mini-SWE-agent | Sep 2026 | Official leaderboard | |
| 19 | 56.9% | xhigh, mini-SWE-agent | Jul 2026 | Official leaderboard | |
| 20 | Claude Haiku 5.5 | 52.9% | high, mini-SWE-agent | Oct 2026 | Official leaderboard |
| 21 | Claude Haiku 4.5 | 25.5% | xhigh, Claude Code | Oct 2025 | Official leaderboard |
Labs report scores for 4 models in their cards. Switch the scores above to rank them together.
Number one over time Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.
Sources: Official SWE-Bench Pro V2 Hard leaderboard and labs’ model cards
Claude Sonnet 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.
Sources: Official SWE-Bench Pro V2 Hard leaderboard and labs’ model cards
Sources: Official SWE-Bench Pro V2 Hard leaderboard and labs’ model cards
3 labs put it in a launch post in the last six months
Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.
| Date | Launch post | Score | As the post writes it |
|---|---|---|---|
| Nov 20, 2026 | 77.9 | SWEBench Pro v2 (Hard), mini-swe-agent v2.4.6 | |
| Nov 5, 2026 | 86.3 | SWEBench Pro v2 (Hard) | |
| Nov 4, 2026 | SWE-2 Flash | 89.9 | SWE-Bench Pro v2 Hard |
| Oct 5, 2026 | Beam 1.0 | 77.2 | SWE Bench Pro v2-Hard |
One system card reports it
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Nov 20, 2026 | | 77.9 | SWEBench Pro v2 (Hard), mini-swe-agent v2.4.6 | Capabilities |
One lab claim next to the leaderboard
Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. 1 of them match, which usually means one side copied the other. Every benchmark.
| Model | Lab says | Board says | Gap |
|---|---|---|---|
| | 86.3% SWEBench Pro v2 (Hard) | 86.3% Leaderboard | Same |
More 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 All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
- Pass Independent runs Scores from the official leaderboard (18 entries), launch posts (4 scores) and system cards (1 score).
- Pass Models scored 21 models we track.
- Pass Newest score The newest is from November 2026, for Inkling 2 Mini.
- Pass Frontier models 4 of the top 10 have a score, led by GPT-6 Astra (#4).
- Note Headroom Saturated. The best is 98%, past 85% since June 2026.
Leaderboard import 3 pass
- Pass Importer We read the board on labs.scale.com: 18 rows for models we track at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 2 pass
- Pass Lab claims The one pair agrees within 10 points.
- Pass Known issues No published audit found problems.