Benchmark Almanac

Snorkel Agentic Coding 2.0

GPT-6 Astra 47.6% leads since October 2026
Not in model cards

Agents plan, run, check and recover on 200 terminal-native engineering tasks across 9 task types and 9 languages. Harbor checks task tests, required outputs, milestones and trace behavior.

Tasks are a frontier subset of Snorkel's Terminal-Bench+ dataset. Scored as Pass@1, unlike Pass@5 in version 1.0.

Made by
Snorkel AI
Released
October 2026
Kind
Building whole projects
Level
Research frontier
Scored
Checked automatically
Answers
Agent task
Tasks
Kept private
Strongest model tested
GPT-6 Astra #3 on the capability index
Last model added
October 2026
Scores refreshed
November 12th, 2026
16
Models scored on its own leaderboard

GPT-6 Astra leads with 47.6%

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.

16 models. Scroll sideways for the rest.

  1. 47.6 GPT-6 Astra
  2. 43.6 Muse Wonder
  3. 42.8 GPT-6.1 Sol
  4. 41.8 Gemini 4 Argon
  5. 40.4 Claude Opus 5.5
  6. 39.6 Claude Fable 5.1
  7. 38.9 Claude Opus 5
  8. 33.6 Grok 4.6
  9. 32.6 Gemini 3.8 Flash
  10. 31.9 Grok 4.7
  11. 26.1 Kimi K3
  12. 25.0 Muse Spark 1.3
  13. 24.9 GLM-5.3
  14. 14.8 DeepSeek-V4-Pro
  15. 14.7 Qwen 3.8 Max
  16. 5.10 Nemotron 3 Ultra
  1. 1 GPT-6 Astra 47.6%
  2. 2 Muse Wonder 43.6%
  3. 3 GPT-6.1 Sol 42.8%
  4. 4 Gemini 4 Argon 41.8%
  5. 5 Claude Opus 5.5 40.4%
  6. 6 Claude Fable 5.1 39.6%
  7. 7 Claude Opus 5 38.9%
  8. 8 Grok 4.6 33.6%
  9. 9 Gemini 3.8 Flash 32.6%
  10. 10 Grok 4.7 31.9%
  11. 11 Kimi K3 26.1%
  12. 12 Muse Spark 1.3 25%
  13. 13 GLM-5.3 24.9%
  14. 14 DeepSeek-V4-Pro 14.8%
  15. 15 Qwen 3.8 Max 14.7%
  16. 16 Nemotron 3 Ultra 5.10%

Source: Official Snorkel Agentic Coding 2.0 leaderboard

#ModelScoreSettingDateSource
1 GPT-6 Astra 47.6% Sep 2026 Official leaderboard
2 Muse Wonder 43.6% Oct 2026 Official leaderboard
3 GPT-6.1 Sol 42.8% Sep 2026 Official leaderboard
4 Gemini 4 Argon 41.8% Sep 2026 Official leaderboard
5 Claude Opus 5.5 40.4% Sep 2026 Official leaderboard
6 Claude Fable 5.1 39.6% Sep 2026 Official leaderboard
7 Claude Opus 5 38.9% Jul 2026 Official leaderboard
8 Grok 4.6 33.6% Aug 2026 Official leaderboard
9 Gemini 3.8 Flash 32.6% Sep 2026 Official leaderboard
10 Grok 4.7 31.9% Sep 2026 Official leaderboard
11 Kimi K3 26.1% Jul 2026 Official leaderboard
12 Muse Spark 1.3 25% Sep 2026 Official leaderboard
13 GLM-5.3 24.9% Aug 2026 Official leaderboard
14 DeepSeek-V4-Pro 14.8% Apr 2026 Official leaderboard
15 Qwen 3.8 Max 14.7% Aug 2026 Official leaderboard
16 Nemotron 3 Ultra 5.10% Jun 2026 Official leaderboard

Pass@1 vs. cost per trial

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% 10% 20% 30% 40% 50% $0$1$2$3$4 Cost per trial ($), lower is better Pass@1 DeepSeek-V4-Pro: Pass@1 14.8%, cost per trial $1.66 Nemotron 3 Ultra: Pass@1 5.1%, cost per trial $0.92 Kimi K3: Pass@1 26.1%, cost per trial $1.58 Claude Opus 5: Pass@1 38.9%, cost per trial $2.91 Qwen 3.8 Max: Pass@1 14.7%, cost per trial $0.9 Grok 4.6: Pass@1 33.6%, cost per trial $1.11 GLM-5.3: Pass@1 24.9%, cost per trial $1.2 Claude Fable 5.1: Pass@1 39.6%, cost per trial $3.86 Gemini 3.8 Flash: Pass@1 32.6%, cost per trial $1.56 Muse Spark 1.3: Pass@1 25%, cost per trial $1.83 Grok 4.7: Pass@1 31.9%, cost per trial $1.61 Claude Opus 5.5: Pass@1 40.4%, cost per trial $0.94 Gemini 4 Argon: Pass@1 41.8%, cost per trial $2.59 Muse Wonder: Pass@1 43.6%, cost per trial $3.31 GPT-6 Astra: Pass@1 47.6%, cost per trial $1.16 GPT-6.1 Sol: Pass@1 42.8%, cost per trial $0.17 GPT-6 AstraGPT-6.1 Sol 0% 10% 20% 30% 40% 50% $0$1$2$3$4 Cost per trial ($), lower is better Pass@1 DeepSeek-V4-Pro: Pass@1 14.8%, cost per trial $1.66 Nemotron 3 Ultra: Pass@1 5.1%, cost per trial $0.92 Kimi K3: Pass@1 26.1%, cost per trial $1.58 Claude Opus 5: Pass@1 38.9%, cost per trial $2.91 Qwen 3.8 Max: Pass@1 14.7%, cost per trial $0.9 Grok 4.6: Pass@1 33.6%, cost per trial $1.11 GLM-5.3: Pass@1 24.9%, cost per trial $1.2 Claude Fable 5.1: Pass@1 39.6%, cost per trial $3.86 Gemini 3.8 Flash: Pass@1 32.6%, cost per trial $1.56 Muse Spark 1.3: Pass@1 25%, cost per trial $1.83 Grok 4.7: Pass@1 31.9%, cost per trial $1.61 Claude Opus 5.5: Pass@1 40.4%, cost per trial $0.94 Gemini 4 Argon: Pass@1 41.8%, cost per trial $2.59 Muse Wonder: Pass@1 43.6%, cost per trial $3.31 GPT-6 Astra: Pass@1 47.6%, cost per trial $1.16 GPT-6.1 Sol: Pass@1 42.8%, cost per trial $0.17 GPT-6 AstraGPT-6.1 Sol
ModelSettingPass@1Cost per trial
GPT-6 Astra 47.6% $1.16
Muse Wonder 43.6% $3.31
GPT-6.1 Sol 42.8% $0.17
Gemini 4 Argon 41.8% $2.59
Claude Opus 5.5 40.4% $0.94
Claude Fable 5.1 39.6% $3.86
Claude Opus 5 38.9% $2.91
Grok 4.6 33.6% $1.11
Gemini 3.8 Flash 32.6% $1.56
Grok 4.7 31.9% $1.61
Kimi K3 26.1% $1.58
Muse Spark 1.3 25% $1.83
GLM-5.3 24.9% $1.2
DeepSeek-V4-Pro 14.8% $1.66
Qwen 3.8 Max 14.7% $0.9
Nemotron 3 Ultra 5.1% $0.92

The best score, 47.6%, costs $1.16 a task

Runs on the cost frontier score higher than any run that's cheaper. GPT-6.1 Sol comes within 4.8 points of the best for $0.17 a task, 7 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Cost frontier One model at each effort level
0% 10% 20% 30% 40% 50% $0.10 $0.20 $0.50 $1.00 $2.00 $5.00 $10 Cost per task, log scale GPT-6 Astra GPT-6.1 Sol 0% 10% 20% 30% 40% 50% $0.10 $1.00 $10 Cost per task, log scale GPT-6 Astra GPT-6.1 Sol

Source: Official Snorkel Agentic Coding 2.0 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6 Astra 47.6% $1.16
Muse Wonder 43.6% $3.31
GPT-6.1 Sol 42.8% $0.17
Gemini 4 Argon 41.8% $2.59
Claude Opus 5.5 40.4% $0.94
Claude Fable 5.1 39.6% $3.86
Claude Opus 5 38.9% $2.91
Grok 4.6 33.6% $1.11
Gemini 3.8 Flash 32.6% $1.56
Grok 4.7 31.9% $1.61
Kimi K3 26.1% $1.58
Muse Spark 1.3 25% $1.83
GLM-5.3 24.9% $1.20
DeepSeek-V4-Pro 14.8% $1.66
Qwen 3.8 Max 14.7% $0.90
Nemotron 3 Ultra 5.10% $0.92

Number one over time

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 47.6%

Source: Official Snorkel Agentic Coding 2.0 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 Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-6 Astra Claude Fable 5.1 Kimi K3 DeepSeek-V4-Pro 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-6 Astra Claude Fable 5.1

Source: Official Snorkel Agentic Coding 2.0 leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 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 OpenAI
Snorkel Agentic Coding 2.0 3 leaders
GPT-6 Astra 47.6%
Snorkel Agentic Coding Not tested since Feb 2026
Claude Opus 4.6 65.2%
Snorkel Agentic Coding 2.0 1 leader
GPT-6 Astra 47.6%

Source: Official leaderboards

Replaces Snorkel Agentic Coding

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