Benchmark Almanac

SWE-sweep

GPT-6 Astra 9.50% leads since October 2026
Gemini 4 Carbon 9.60% leads since October 2026
In use

Agents get a real repository with no hint of bug type or location and must find and repair as many bugs as they can. Score is the share of bugs resolved.

100 repositories with 4.1k bugs, run through Harbor with mini-SWE-agent. Authors include the SWE-bench team.

Released
October 2026
Kind
Fixing real repositories
Level
Research frontier
Scored
Checked automatically
Answers
Code
Tasks
All public
Strongest model tested
Gemini 4 Carbon #2 on the capability index
Last in a launch post
October 2026
Last model added
October 2026
Scores refreshed
November 12th, 2026
10
Models scored on its own leaderboard
1
Launch post lists it, from 1 lab

GPT-6 Astra leads with 9.50% among models released by September 2026

Best score per model from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

Gemini 4 Carbon leads with 9.60%, 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.

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:

  1. 9.50 GPT-6 Astra
  2. 4.70 GPT-5.6 Sol
  3. 3.60 Claude Opus 5.5
  4. 2.90 Claude Fable 5.1
  5. 2.50 GPT-5.6 Luna
  6. 1.50 GPT-5.6 Terra
  7. 1.30 Claude Opus 5
  8. 0.60 Kimi K3
  9. 0.50 GPT-5.4 Mini
  10. 0.10 Gemini 3.5 Flash-Lite
  1. 1 GPT-6 Astra 9.50%
  2. 2 GPT-5.6 Sol xhigh 4.70%
  3. 3 Claude Opus 5.5 xhigh 3.60%
  4. 4 Claude Fable 5.1 2.90%
  5. 5 GPT-5.6 Luna xhigh 2.50%
  6. 6 GPT-5.6 Terra xhigh 1.50%
  7. 7 Claude Opus 5 xhigh 1.30%
  8. 8 Kimi K3 0.60%
  9. 9 GPT-5.4 Mini high 0.50%
  10. 10 Gemini 3.5 Flash-Lite 0.10%

Source: Official SWE-sweep leaderboard

Sources: Labs’ model cards and official SWE-sweep leaderboard

#ModelScoreSettingDateSource
1 GPT-6 Astra 9.50% mini-SWE-agent Sep 2026 Official leaderboard
2 GPT-5.6 Sol 4.70% xhigh, mini-SWE-agent Jul 2026 Official leaderboard
3 Claude Opus 5.5 3.60% xhigh, mini-SWE-agent Sep 2026 Official leaderboard
4 Claude Fable 5.1 2.90% mini-SWE-agent Sep 2026 Official leaderboard
5 GPT-5.6 Luna 2.50% xhigh, mini-SWE-agent Jul 2026 Official leaderboard
6 GPT-5.6 Terra 1.50% xhigh, mini-SWE-agent Jul 2026 Official leaderboard
7 Claude Opus 5 1.30% xhigh, mini-SWE-agent Jul 2026 Official leaderboard
8 Kimi K3 0.60% mini-SWE-agent Jul 2026 Official leaderboard
9 GPT-5.4 Mini 0.50% high, mini-SWE-agent Mar 2026 Official leaderboard
10 Gemini 3.5 Flash-Lite 0.10% mini-SWE-agent Jul 2026 Official leaderboard
#ModelScoreSettingDateSource
1 Gemini 4 Carbon 9.60% Oct 2026 Lab model card
2 GPT-6 Astra 9.50% mini-SWE-agent Sep 2026 Official leaderboard
3 GPT-5.6 Sol 4.70% xhigh, mini-SWE-agent Jul 2026 Official leaderboard
4 Claude Opus 5.5 3.60% xhigh, mini-SWE-agent Sep 2026 Official leaderboard
5 Claude Fable 5.1 2.90% mini-SWE-agent Sep 2026 Official leaderboard
6 GPT-5.6 Luna 2.50% xhigh, mini-SWE-agent Jul 2026 Official leaderboard
7 GPT-5.6 Terra 1.50% xhigh, mini-SWE-agent Jul 2026 Official leaderboard
8 Claude Opus 5 1.30% xhigh, mini-SWE-agent Jul 2026 Official leaderboard
9 Kimi K3 0.60% mini-SWE-agent Jul 2026 Official leaderboard
10 GPT-5.4 Mini 0.50% high, mini-SWE-agent Mar 2026 Official leaderboard
11 Gemini 3.5 Flash-Lite 0.10% mini-SWE-agent Jul 2026 Official leaderboard

Labs report scores for 1 model in their cards. Switch the scores above to rank them together.

Pareto: bugs resolved against turns

The Pareto tab with turns on the x axis: each point is one model, and the line marks the best result for a given number of turns per repository. 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% 2% 4% 6% 8% 10% 0100200300400 Turns per repository, lower is better Bugs resolved GPT-5.4 Mini (high): Bugs resolved 0.5%, turns per repository 75 GPT-5.6 Luna (xhigh): Bugs resolved 2.5%, turns per repository 204 GPT-5.6 Sol (xhigh): Bugs resolved 4.7%, turns per repository 233 Kimi K3: Bugs resolved 0.6%, turns per repository 337 Gemini 3.5 Flash-Lite: Bugs resolved 0.1%, turns per repository 34 Claude Opus 5 (xhigh): Bugs resolved 1.3%, turns per repository 323 Claude Fable 5.1: Bugs resolved 2.9%, turns per repository 319 Claude Opus 5.5 (xhigh): Bugs resolved 3.6%, turns per repository 266 GPT-5.4 Mini: Bugs resolved 0.2%, turns per repository 13 GPT-5.6 Luna (high): Bugs resolved 1.4%, turns per repository 75 GPT-5.6 Luna: Bugs resolved 0.5%, turns per repository 22 GPT-5.6 Terra (xhigh): Bugs resolved 1.5%, turns per repository 76 GPT-6 Astra: Bugs resolved 9.5%, turns per repository 113 GPT-6 AstraGPT-5.6 Terra (xhigh)GPT-5.6 Luna (high)GPT-5.6 LunaGPT-5.4 Mini 0% 2% 4% 6% 8% 10% 0100200300400 Turns per repository, lower is better Bugs resolved GPT-5.4 Mini (high): Bugs resolved 0.5%, turns per repository 75 GPT-5.6 Luna (xhigh): Bugs resolved 2.5%, turns per repository 204 GPT-5.6 Sol (xhigh): Bugs resolved 4.7%, turns per repository 233 Kimi K3: Bugs resolved 0.6%, turns per repository 337 Gemini 3.5 Flash-Lite: Bugs resolved 0.1%, turns per repository 34 Claude Opus 5 (xhigh): Bugs resolved 1.3%, turns per repository 323 Claude Fable 5.1: Bugs resolved 2.9%, turns per repository 319 Claude Opus 5.5 (xhigh): Bugs resolved 3.6%, turns per repository 266 GPT-5.4 Mini: Bugs resolved 0.2%, turns per repository 13 GPT-5.6 Luna (high): Bugs resolved 1.4%, turns per repository 75 GPT-5.6 Luna: Bugs resolved 0.5%, turns per repository 22 GPT-5.6 Terra (xhigh): Bugs resolved 1.5%, turns per repository 76 GPT-6 Astra: Bugs resolved 9.5%, turns per repository 113 GPT-6 AstraGPT-5.6 Terra (xhigh)GPT-5.6 Luna (high)GPT-5.6 LunaGPT-5.4 Mini
ModelSettingBugs resolvedTurns per repository
GPT-6 Astra mini-SWE-agent 9.5% 113
GPT-5.6 Sol xhigh, mini-SWE-agent 4.7% 233
Claude Opus 5.5 xhigh, mini-SWE-agent 3.6% 266
Claude Fable 5.1 mini-SWE-agent 2.9% 319
GPT-5.6 Luna xhigh, mini-SWE-agent 2.5% 204
GPT-5.6 Terra xhigh, mini-SWE-agent 1.5% 76
GPT-5.6 Luna high, mini-SWE-agent 1.4% 75
Claude Opus 5 xhigh, mini-SWE-agent 1.3% 323
Kimi K3 mini-SWE-agent 0.6% 337
GPT-5.6 Luna mini-SWE-agent 0.5% 22
GPT-5.4 Mini high, mini-SWE-agent 0.5% 75
GPT-5.4 Mini mini-SWE-agent 0.2% 13
Gemini 3.5 Flash-Lite mini-SWE-agent 0.1% 34

The best score, 9.50%, costs $35 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.

OpenAI Anthropic Cost frontier One model at each effort level
0% 2% 4% 6% 8% 10% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6 Astra Claude Opus 5.5, xhigh GPT-5.6 Luna, xhigh GPT-5.6 Luna, high GPT-5.6 Luna 0% 2% 4% 6% 8% 10% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6 Astra Claude Opus 5.5, xhigh

Source: Official SWE-sweep leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6 Astra mini-SWE-agent 9.50% $35 119,500
GPT-5.6 Sol xhigh mini-SWE-agent 4.70% $72 104,800
Claude Opus 5.5 xhigh mini-SWE-agent 3.60% $22 112,700
Claude Fable 5.1 mini-SWE-agent 2.90% $39 165,000
GPT-5.6 Luna xhigh mini-SWE-agent 2.50% $2.24 61,200
GPT-5.6 Terra xhigh mini-SWE-agent 1.50% $3.57 46,600
GPT-5.6 Luna high mini-SWE-agent 1.40% $0.28 20,300
Claude Opus 5 xhigh mini-SWE-agent 1.30% $54 192,700
Kimi K3 mini-SWE-agent 0.60% $25 152,600
GPT-5.4 Mini high mini-SWE-agent 0.50% $1.22 40,500
GPT-5.6 Luna mini-SWE-agent 0.50% $0.04 4,500
GPT-5.4 Mini mini-SWE-agent 0.20% $0.05 2,100
Gemini 3.5 Flash-Lite mini-SWE-agent 0.10% $0.06 5,200

Number one over time 2 models have held number one with claims included since its release in October 2026

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

OpenAI
Own leaderboard 1 leader
GPT-6 Astra 9.50%

Source: Official SWE-sweep 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 Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Gemini 4 Carbon GPT-5.6 Sol GPT-5.4 Mini 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Gemini 4 Carbon GPT-5.4 Mini

Sources: Official SWE-sweep leaderboard and labs’ model cards

One lab 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.

Google DeepMind
DateLaunch postScoreAs the post writes it
Oct 28, 2026 Gemini 4 Carbon 9.6 SWE-sweep

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