Benchmark Almanac

FrontierSWE v1

Claude Fable 5 88.2% leads since June 2026
Claude Fable 5 88.2% leads since June 2026
In use

Agents work up to 20 hours on 17 ultra-long engineering, performance and research tasks in their own harnesses. The board ranks models by average task rank. The score here is dominance, the win rate against a random baseline, which labs report.

The maker preserves the v1 board as published. FrontierSWE v2 replaced it on September 2nd, 2026 with 34 tasks, a new harness and mean@5 scores.

Made by
Proximal
Released
April 2026
Kind
Building whole projects
Level
Research frontier
Scored
Checked automatically
Answers
Code
Tasks
All public
Strongest model tested
Claude Fable 5 #11 on the capability index
Last in a launch post
August 2026
Last model added
August 2026
Scores refreshed
November 12th, 2026
17
Models scored on its own leaderboard
4
Launch posts list it, from 3 labs
4
System cards report it, from 3 labs

Claude Fable 5 leads with 88.2% among models released by August 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.

Claude Fable 5 leads with 88.2%, 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 August 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

17 models. Scroll sideways for the rest.

  1. 88.2 Claude Fable 5
  2. 78.1 GLM-5.3
  3. 77.9 Grok 4.6
  4. 72.1 Grok 4.5
  5. 67.5 GLM-5.2
  6. 66.5 Claude Opus 4.8
  7. 64.5 GPT-5.5
  8. 56.3 Claude Opus 4.7
  9. 48.9 Claude Opus 4.6
  10. 46.0 GPT-5.4
  11. 34.4 Gemini 3.1 Pro
  12. 33.8 Composer 2.5
  13. 25.7 GLM-5.1
  14. 24.6 DeepSeek-V4-Pro
  15. 23.4 Kimi K2.5
  16. 22.2 Kimi K2.6
  17. 19.9 Qwen 3.6 Plus
  1. 1 Claude Fable 5 mean 88.2%
  2. 2 GLM-5.3 mean 78.1%
  3. 3 Grok 4.6 mean 77.9%
  4. 4 Grok 4.5 mean 72.1%
  5. 5 GLM-5.2 mean 67.5%
  6. 6 Claude Opus 4.8 mean 66.5%
  7. 7 GPT-5.5 mean 64.5%
  8. 8 Claude Opus 4.7 mean 56.3%
  9. 9 Claude Opus 4.6 mean 48.9%
  10. 10 GPT-5.4 mean 46%
  11. 11 Gemini 3.1 Pro mean 34.4%
  12. 12 Composer 2.5 mean 33.8%
  13. 13 GLM-5.1 mean 25.7%
  14. 14 DeepSeek-V4-Pro mean 24.6%
  15. 15 Kimi K2.5 mean 23.4%
  16. 16 Kimi K2.6 mean 22.2%
  17. 17 Qwen 3.6 Plus mean 19.9%

Source: Official FrontierSWE v1 leaderboard

19 models. Scroll sideways for the rest.

  1. 88.2 Claude Fable 5
  2. 81.2 Kimi K3
  3. 78.1 GLM-5.3
  4. 78.0 Grok 4.5
  5. 77.9 Grok 4.6
  6. 74.4 GLM-5.2
  7. 73.5 Qwen 3.8 Max
  8. 66.5 Claude Opus 4.8
  9. 64.5 GPT-5.5
  10. 56.3 Claude Opus 4.7
  11. 48.9 Claude Opus 4.6
  12. 46.0 GPT-5.4
  13. 34.4 Gemini 3.1 Pro
  14. 33.8 Composer 2.5
  15. 25.7 GLM-5.1
  16. 24.6 DeepSeek-V4-Pro
  17. 23.4 Kimi K2.5
  18. 22.2 Kimi K2.6
  19. 19.9 Qwen 3.6 Plus
  1. 1 Claude Fable 5 mean 88.2%
  2. 2 Kimi K3 81.2%
  3. 3 GLM-5.3 mean 78.1%
  4. 4 Grok 4.5 78%
  5. 5 Grok 4.6 mean 77.9%
  6. 6 GLM-5.2 74.4%
  7. 7 Qwen 3.8 Max 73.5%
  8. 8 Claude Opus 4.8 mean 66.5%
  9. 9 GPT-5.5 mean 64.5%
  10. 10 Claude Opus 4.7 mean 56.3%
  11. 11 Claude Opus 4.6 mean 48.9%
  12. 12 GPT-5.4 mean 46%
  13. 13 Gemini 3.1 Pro mean 34.4%
  14. 14 Composer 2.5 mean 33.8%
  15. 15 GLM-5.1 mean 25.7%
  16. 16 DeepSeek-V4-Pro mean 24.6%
  17. 17 Kimi K2.5 mean 23.4%
  18. 18 Kimi K2.6 mean 22.2%
  19. 19 Qwen 3.6 Plus mean 19.9%

Sources: Official FrontierSWE v1 leaderboard and labs’ model cards

#ModelScoreSettingDateSource
1 Claude Fable 5 88.2% mean, Claude Code Jun 2026 Official leaderboard
2 GLM-5.3 78.1% mean, Claude Code Aug 2026 Official leaderboard
3 Grok 4.6 77.9% mean, Grok CLI Aug 2026 Official leaderboard
4 Grok 4.5 72.1% mean, Grok CLI Jul 2026 Official leaderboard
5 GLM-5.2 67.5% mean, Claude Code Jun 2026 Official leaderboard
6 Claude Opus 4.8 66.5% mean, Claude Code May 2026 Official leaderboard
7 GPT-5.5 64.5% mean, Codex Apr 2026 Official leaderboard
8 Claude Opus 4.7 56.3% mean, Claude Code Apr 2026 Official leaderboard
9 Claude Opus 4.6 48.9% mean, Claude Code Feb 2026 Official leaderboard
10 GPT-5.4 46% mean, Codex Mar 2026 Official leaderboard
11 Gemini 3.1 Pro 34.4% mean, Gemini CLI Feb 2026 Official leaderboard
12 Composer 2.5 33.8% mean, Cursor CLI May 2026 Official leaderboard
13 GLM-5.1 25.7% mean, Claude Code Apr 2026 Official leaderboard
14 DeepSeek-V4-Pro 24.6% mean, Claude Code Apr 2026 Official leaderboard
15 Kimi K2.5 23.4% mean, Kimi CLI Jan 2026 Official leaderboard
16 Kimi K2.6 22.2% mean, Kimi CLI Apr 2026 Official leaderboard
17 Qwen 3.6 Plus 19.9% mean, Qwen Code Mar 2026 Official leaderboard
#ModelScoreSettingDateSource
1 Claude Fable 5 88.2% mean, Claude Code Jun 2026 Official leaderboard
2 Kimi K3 81.2% Jul 2026 Lab model card
3 GLM-5.3 78.1% mean, Claude Code Aug 2026 Official leaderboard
4 Grok 4.5 78% Jul 2026 Lab model card
5 Grok 4.6 77.9% mean, Grok CLI Aug 2026 Official leaderboard
6 GLM-5.2 74.4% Jun 2026 Lab model card
7 Qwen 3.8 Max 73.5% Aug 2026 Lab model card
8 Claude Opus 4.8 66.5% mean, Claude Code May 2026 Official leaderboard
9 GPT-5.5 64.5% mean, Codex Apr 2026 Official leaderboard
10 Claude Opus 4.7 56.3% mean, Claude Code Apr 2026 Official leaderboard
11 Claude Opus 4.6 48.9% mean, Claude Code Feb 2026 Official leaderboard
12 GPT-5.4 46% mean, Codex Mar 2026 Official leaderboard
13 Gemini 3.1 Pro 34.4% mean, Gemini CLI Feb 2026 Official leaderboard
14 Composer 2.5 33.8% mean, Cursor CLI May 2026 Official leaderboard
15 GLM-5.1 25.7% mean, Claude Code Apr 2026 Official leaderboard
16 DeepSeek-V4-Pro 24.6% mean, Claude Code Apr 2026 Official leaderboard
17 Kimi K2.5 23.4% mean, Kimi CLI Jan 2026 Official leaderboard
18 Kimi K2.6 22.2% mean, Kimi CLI Apr 2026 Official leaderboard
19 Qwen 3.6 Plus 19.9% mean, Qwen Code Mar 2026 Official leaderboard

Labs report scores for 5 models in their cards. Switch the scores above to rank them together.

5 models have held number one on the official leaderboard since its release in April 2026 5 models have held number one with claims included since its release in April 2026

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.

Anthropic OpenAI Z.ai xAI Moonshot AI
Own leaderboard Passed 85% in Jun 2026, by Claude Fable 5
Claude Fable 5 88.2%
Lab claims Replaced by FrontierSWE, Sep 2026
Kimi K3 81.2%

Sources: Official FrontierSWE v1 leaderboard and labs’ model cards

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

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 Near the ceiling (85%) Claude Fable 5 Claude Opus 4.8 Claude Opus 4.7 Claude Opus 4.6 Kimi K2.5 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Fable 5 Claude Opus 4.8

Sources: Official FrontierSWE v1 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.

Z.ai
Moonshot AI
Qwen
DateLaunch postScoreAs the post writes it
Aug 14, 2026 GLM-5.3 78.1 FrontierSWE
Aug 3, 2026 Qwen 3.8 Max 73.5 FrontierSWE
Jul 16, 2026 Kimi K3 81.2 FrontierSWE
Jun 16, 2026 GLM-5.2 74.4 FrontierSWE (Dominance)

4 system cards report it, from 3 labs

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.

DateDocumentScoreAs the document writes itSection
Jul 27, 2026 Kimi K3 technical report 81.2 FrontierSWE dominance score recomputed with the official script as of 2026-07-16 (Table 2) Capabilities
Jul 14, 2026 Grok 4.5 model card 78 FrontierSWE, Dominance (win rate versus a random baseline), high, Grok CLI, public leaderboard (2.7) Capabilities
Jun 9, 2026 Claude Fable 5 system card FrontierSWE, 17 tasks, ranked #1 on mean@5 with an average rank of 2.12 (lower is better), no mean task score printed (8.5) Capabilities
May 28, 2026 Claude Opus 4.8 system card FrontierSWE leaderboard at xhigh effort, ranked first on mean@5 (average rank 2.74) and best@5 (average rank 2.26), no score printed (8.4) Capabilities

The lab's number comes out ahead in 2 of 2 pairs that differ

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
GLM-5.2 74.4% FrontierSWE (Dominance) 67.5% Leaderboard, mean, Claude Code +6.9 pts
Grok 4.5 78% FrontierSWE, Dominance (win rate versus a random baseline), high, Grok CLI, public leaderboard (2.7) 72.1% Leaderboard, mean, Grok CLI +5.9 pts
GLM-5.3 78.1% FrontierSWE 78.1% Leaderboard, mean, Claude Code Same

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 OpenAI
FrontierSWE 8 leaders
Claude Fable 5.2 69.9%
FrontierSWE v1 Passed 85% in Jun 2026, by Claude Fable 5
Claude Fable 5 88.2%
FrontierSWE 3 leaders
Claude Fable 5.2 69.9%

Source: Official leaderboards

Replaced by FrontierSWE

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