Benchmark Almanac

SWE-Serve

GPT-5.6 Sol 75.5% leads since October 2026
Not in model cards

Coding agents solve 53 production inference-engineering tasks derived from merged SGLang pull requests. Hidden tests measure task success, cost, agent steps and wall time.

NVIDIA's benchmark page lists 53 tasks across six engineering areas, with 41 single-H100 tasks and 12 CPU tasks.

Made by
NVIDIA
Released
October 2026
Kind
Fixing real repositories
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
Public, plus a private test set
Measures
Capability
Language
English
In and out
Text, Code → Code, Actions
Interaction
Agentic
Environment
Terminal
Items from
Real work and use
Refresh
One fixed set
Run by
A host
Occupations
Computer and mathematical
In views
Agents
Strongest model tested
Claude Opus 5 #11 on the capability index
Last model added
July 2026
Scores refreshed
November 21st, 2026
11
Entries on its own leaderboard, 10 models we track

2 models share the top score of 75.5% among models released by July 2026 2 models share the top score of 75.5% among models released by July 2026

Best score per entry from the benchmark's own leaderboard, read on November 21st, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.

The newest model with a score here came out in July 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT-5.6 Sol max 75.5%
  2. 2 Claude Opus 5 max 75.5%
  3. 3 Claude Sonnet 5 xhigh 64.2%
  4. 4 GPT-5.6 Luna max 64.2%
  5. 5 GPT-5.6 Terra max 64.2%
  6. 6 Kimi K3 max 64.2%
  7. 7 DeepSeek V4 Flash 0731 xhigh 55.4%
  8. 8 GLM-5.2 max 48.4%
  9. 9 Gemini 3.6 Flash high 47.8%
  10. 10 Laguna S 2.1 xhigh 45.9%
  11. 11 Inkling S xhigh 34.6%

Source: Official SWE-Serve leaderboard

#EntryScore ±SettingDateSource
1 GPT-5.6 Sol 75.5% ±6.4 max Jul 2026 Official leaderboard
2 Claude Opus 5 75.5% ±3.7 max Jul 2026 Official leaderboard
3 Claude Sonnet 5 64.2% ±3.7 xhigh Jun 2026 Official leaderboard
4 GPT-5.6 Luna 64.2% ±4.3 max Jul 2026 Official leaderboard
5 GPT-5.6 Terra 64.2% ±4.3 max Jul 2026 Official leaderboard
6 Kimi K3 64.2% ±5.6 max Jul 2026 Official leaderboard
7 DeepSeek V4 Flash 0731 55.4% ±4.9 xhigh Jul 2026 Official leaderboard
8 GLM-5.2 48.4% ±2.5 max Jun 2026 Official leaderboard
9 Gemini 3.6 Flash 47.8% ±6.5 high Jul 2026 Official leaderboard
10 Laguna S 2.1 45.9% ±5.4 xhigh Jul 2026 Official leaderboard
11 Inkling S 34.6% ±3.3 xhigh Official leaderboard
#ModelScore ±SettingDateSource
1 GPT-5.6 Sol 75.5% ±6.4 max Jul 2026 Official leaderboard
2 Claude Opus 5 75.5% ±3.7 max Jul 2026 Official leaderboard
3 Claude Sonnet 5 64.2% ±3.7 xhigh Jun 2026 Official leaderboard
4 GPT-5.6 Luna 64.2% ±4.3 max Jul 2026 Official leaderboard
5 GPT-5.6 Terra 64.2% ±4.3 max Jul 2026 Official leaderboard
6 Kimi K3 64.2% ±5.6 max Jul 2026 Official leaderboard
7 DeepSeek V4 Flash 0731 55.4% ±4.9 xhigh Jul 2026 Official leaderboard
8 GLM-5.2 48.4% ±2.5 max Jun 2026 Official leaderboard
9 Gemini 3.6 Flash 47.8% ±6.5 high Jul 2026 Official leaderboard
10 Laguna S 2.1 45.9% ±5.4 xhigh Jul 2026 Official leaderboard

The best score, 75.5%, costs $12 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 Poolside Anthropic Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale GPT-5.6 Sol, max Claude Opus 5, medium GPT-5.6 Luna, max GPT-5.6 Luna, xhigh Laguna S 2.1, xhigh GPT-5.6 Luna, medium GPT-5.6 Luna, low 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale GPT-5.6 Sol, max Claude Opus 5, medium

Source: Official SWE-Serve leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.6 Sol max 75.5% $12 51,800
Claude Opus 5 max 75.5% $17 122,016
Claude Opus 5 high 73.6% $9.89 69,380
Claude Opus 5 xhigh 72.3% $14 96,689
Claude Opus 5 medium 68.5% $5.08 39,237
GPT-5.6 Sol xhigh 66% $6.04 34,716
Claude Sonnet 5 xhigh 64.2% $6.61 98,444
GPT-5.6 Luna max 64.2% $0.95 64,591
GPT-5.6 Terra max 64.2% $5.06 60,838
Kimi K3 max 64.2% $7.24 79,340
Claude Sonnet 5 max 61% $15 198,103
Claude Opus 5 low 60.4% $1.96 17,679
Claude Sonnet 5 high 57.9% $4.54 71,561
GPT-5.6 Sol high 56% $3.22 24,436
Claude Sonnet 5 medium 55.4% $2.38 41,505
GPT-5.6 Luna xhigh 55.4% $0.36 39,416
DeepSeek V4 Flash 0731 xhigh 55.4% $0.69 98,515
GPT-5.6 Sol medium 54.1% $1.69 14,086
GPT-5.6 Terra xhigh 52.8% $1.45 31,354
Claude Sonnet 5 low 49.1% $0.86 19,527
GPT-5.6 Terra high 49.1% $0.67 16,501
GLM-5.2 max 48.4% $2.10 53,304
Gemini 3.6 Flash high 47.8% $4.84 99,465
Laguna S 2.1 xhigh 45.9% $0.33 146,737
GPT-5.6 Luna high 45.3% $0.12 20,809
GPT-5.6 Terra medium 44.6% $0.33 8,941
GPT-5.6 Terra low 44% $0.25 6,873
GPT-5.6 Sol low 40.9% $0.61 6,525
GPT-5.6 Luna medium 37.7% $0.02 5,791
GPT-5.6 Luna low 20.8% $0.01 2,718

The best score, 75.5%, costs $12 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 Poolside Anthropic Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale GPT-5.6 Sol, max Claude Opus 5, medium GPT-5.6 Luna, max GPT-5.6 Luna, xhigh Laguna S 2.1, xhigh GPT-5.6 Luna, medium GPT-5.6 Luna, low 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale GPT-5.6 Sol, max Claude Opus 5, medium

Source: Official SWE-Serve leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.6 Sol max 75.5% $12 51,800
Claude Opus 5 max 75.5% $17 122,016
Claude Opus 5 high 73.6% $9.89 69,380
Claude Opus 5 xhigh 72.3% $14 96,689
Claude Opus 5 medium 68.5% $5.08 39,237
GPT-5.6 Sol xhigh 66% $6.04 34,716
Claude Sonnet 5 xhigh 64.2% $6.61 98,444
GPT-5.6 Luna max 64.2% $0.95 64,591
GPT-5.6 Terra max 64.2% $5.06 60,838
Kimi K3 max 64.2% $7.24 79,340
Claude Sonnet 5 max 61% $15 198,103
Claude Opus 5 low 60.4% $1.96 17,679
Claude Sonnet 5 high 57.9% $4.54 71,561
GPT-5.6 Sol high 56% $3.22 24,436
Claude Sonnet 5 medium 55.4% $2.38 41,505
GPT-5.6 Luna xhigh 55.4% $0.36 39,416
DeepSeek V4 Flash 0731 xhigh 55.4% $0.69 98,515
GPT-5.6 Sol medium 54.1% $1.69 14,086
GPT-5.6 Terra xhigh 52.8% $1.45 31,354
Claude Sonnet 5 low 49.1% $0.86 19,527
GPT-5.6 Terra high 49.1% $0.67 16,501
GLM-5.2 max 48.4% $2.10 53,304
Gemini 3.6 Flash high 47.8% $4.84 99,465
Laguna S 2.1 xhigh 45.9% $0.33 146,737
GPT-5.6 Luna high 45.3% $0.12 20,809
GPT-5.6 Terra medium 44.6% $0.33 8,941
GPT-5.6 Terra low 44% $0.25 6,873
GPT-5.6 Sol low 40.9% $0.61 6,525
GPT-5.6 Luna medium 37.7% $0.02 5,791
GPT-5.6 Luna low 20.8% $0.01 2,718

Number one over time

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

OpenAI
Own leaderboard Not tested since Oct 2026
GPT-5.6 Sol 75.5%

Source: Official SWE-Serve 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% 2026 Near the ceiling (85%) GPT-5.6 Sol Claude Sonnet 5 GLM-5.2 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-5.6 Sol Claude Sonnet 5

Source: Official SWE-Serve 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.

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 10 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access Public tasks, plus a private test set.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (11 entries).
  • Pass Models scored 10 models we track, plus 1 other entry on its leaderboard.
  • Pass Newest score The newest is from July 2026, for DeepSeek V4 Flash 0731.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Claude Opus 5 (#11).
  • Pass Headroom The best is 75.5%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on research.nvidia.com: 30 rows for models we track and 1 other entry at the last read.
  • Pass Last read November 21st, 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