Benchmark Almanac

JobBench

GLM-5.3 61.4% leads since August 2026
Muse Wonder Full 75.3% leads since October 2026
In use

The official leaderboard hasn’t scored a model released after GLM-5.3 (August 2026). Labs report their own numbers for 14 newer models, such as Muse Wonder Full at 75.3%.

130 tasks from 35 occupations workers most want delegated. Agent works through messy reference files, graded by fact-anchored binary rubrics.

Framed as an alternative to GDPval's economic-value framing. Vendor-reported scores run higher than the paper's best (45.9%).

Released
May 2026
Kind
Tasks from real jobs
Level
Expert
Scored
Graded by a model
Answers
Open-ended
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Documents, Tables → Actions, Documents, Text
Interaction
Agentic
Environment
Terminal
Refresh
One fixed set
Run by
Anyone
Strongest model tested
Claude Fable 5 #12 on the capability index
Last in a launch post
November 2026
Last model added
November 2026
Scores refreshed
November 19th, 2026

In the authors’ words

Current benchmarks for occupational AI agents are scoped primarily by economic values, telling a replacement story. We introduce JobBench, which evaluates AI agents on the workflows that experts identify as high-priority for delegation, empowering humans based on their needs instead of replacing them with GDP value. JobBench covers 130 agentic tasks across 35 occupations. Each task is packaged as a workspace of heterogeneous reference files, requiring the agent to reason through the cluttered information streams of real professional work. Outputs are graded by a fact-anchored chain of rubrics, averaging 35.6 binary criteria per task. We evaluate 36 models; the strongest, Claude Opus 4.7 under Claude Code, reaches only 45.9 %. We hope JobBench shifts the community's target labour-market effect from replacement to enhancement: building agents that do what humans actually want delegated, not only what is most economically valuable.

Abstract of JobBench: Aligning Agent Work With Human Will (arXiv:2605.26329, CC0)
16
Models scored on its own leaderboard
18
Launch posts list it, from 8 labs
6
System cards report it, from 4 labs

GLM-5.3 leads with 61.4% among models released by August 2026

Best score per model from the benchmark's own leaderboard, read on November 19th, 2026. Where a model ran with several settings or harnesses, we keep its best.

Muse Wonder Full leads with 75.3%, 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.

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:

  1. 1 GLM-5.3 61.4%
  2. 2 Claude Fable 5 Claude Opus 4.8 fallback on refusals 57.4%
  3. 3 Muse Spark 1.1 54.7%
  4. 4 Kimi K3 54.3%
  5. 5 Qwen 3.8 Max 52.7%
  6. 6 Claude Opus 4.8 48.4%
  7. 7 GPT-5.6 Sol 45.4%
  8. 8 Claude Opus 4.7 44.5%
  9. 9 GLM-5.2 43.4%
  10. 10 GPT-5.5 38.3%
  11. 11 Claude Sonnet 4.6 36.6%
  12. 12 Gemini 3.5 Flash 31.5%
  13. 13 Inkling 31.1%
  14. 14 Nemotron 3 Ultra 27.2%
  15. 15 Claude Sonnet 4.5 20.7%
  16. 16 Gemini 3.1 Pro 15.9%

Source: Official JobBench leaderboard

#ModelScoreSettingDateSource
1 GLM-5.3 61.4% OpenCode Aug 2026 Official leaderboard
2 Claude Fable 5 57.4% Claude Opus 4.8 fallback on refusals, OpenCode Jun 2026 Official leaderboard
3 Muse Spark 1.1 54.7% OpenCode Jul 2026 Official leaderboard
4 Kimi K3 54.3% OpenCode Jul 2026 Official leaderboard
5 Qwen 3.8 Max 52.7% OpenCode Aug 2026 Official leaderboard
6 Claude Opus 4.8 48.4% OpenCode May 2026 Official leaderboard
7 GPT-5.6 Sol 45.4% OpenCode Jul 2026 Official leaderboard
8 Claude Opus 4.7 44.5% OpenCode Apr 2026 Official leaderboard
9 GLM-5.2 43.4% OpenCode Jun 2026 Official leaderboard
10 GPT-5.5 38.3% OpenCode Apr 2026 Official leaderboard
11 Claude Sonnet 4.6 36.6% OpenCode Feb 2026 Official leaderboard
12 Gemini 3.5 Flash 31.5% OpenCode May 2026 Official leaderboard
13 Inkling 31.1% OpenCode Jul 2026 Official leaderboard
14 Nemotron 3 Ultra 27.2% OpenCode Jun 2026 Official leaderboard
15 Claude Sonnet 4.5 20.7% OpenCode Sep 2025 Official leaderboard
16 Gemini 3.1 Pro 15.9% OpenCode Feb 2026 Official leaderboard
#ModelScoreSettingDateSource
1 Muse Wonder Full 75.3% Oct 2026 Lab model card
2 Muse Wonder 70.5% Oct 2026 Lab model card
3 Inkling 2 Small 70.5% Nov 2026 Lab model card
4 Qwen4-Max 68.5% Nov 2026 Lab model card
5 LongCat-2.5 65.4% Nov 2026 Lab model card
6 Muse Spark 1.3 64.9% Sep 2026 Lab model card
7 Qwen3.8 Max (0902) 64% Sep 2026 Lab model card
8 MiMo-V2.6 Pro 62% Sep 2026 Lab model card
9 Muse Spark Open 61.8% Oct 2026 Lab model card
10 Tencent Hy4 preview 61.7% Aug 2026 Lab model card
11 GLM-5.3 61.4% OpenCode Aug 2026 Official leaderboard
12 MiMo-V2.6 Flash 61.2% Sep 2026 Lab model card
13 Step 5 Preview 59% Sep 2026 Lab model card
14 Claude Fable 5 57.4% Claude Opus 4.8 fallback on refusals, OpenCode Jun 2026 Official leaderboard
15 Qwen3.8 Flash 55.7% Aug 2026 Lab model card
16 Muse Spark 1.1 54.7% OpenCode Jul 2026 Official leaderboard
17 Kimi K3 54.3% OpenCode Jul 2026 Official leaderboard
18 Qwen 3.8 Max 53.4% Aug 2026 Lab model card
19 Inkling 2 Mini 49.6% Nov 2026 Lab model card
20 Claude Opus 4.8 48.4% OpenCode May 2026 Official leaderboard
21 GPT-5.6 Sol 45.4% OpenCode Jul 2026 Official leaderboard
22 Claude Opus 4.7 44.5% OpenCode Apr 2026 Official leaderboard
23 GLM-5.2 43.4% OpenCode Jun 2026 Official leaderboard
24 GPT-5.5 38.3% OpenCode Apr 2026 Official leaderboard
25 Claude Sonnet 4.6 36.6% OpenCode Feb 2026 Official leaderboard
26 Qwen 3.8 27B 33.4% Aug 2026 Lab model card
27 Gemini 3.5 Flash 31.5% OpenCode May 2026 Official leaderboard
28 Inkling 31.1% OpenCode Jul 2026 Official leaderboard
29 Nemotron 3 Ultra 27.2% OpenCode Jun 2026 Official leaderboard
30 Claude Sonnet 4.5 20.7% OpenCode Sep 2025 Official leaderboard
31 Gemini 3.1 Pro 15.9% OpenCode Feb 2026 Official leaderboard

Labs report scores for 18 models in their cards. Switch the scores above to rank them together.

4 models have held number one on the official leaderboard since its release in May 2026 7 models have held number one with claims included since its release in May 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 Z.ai Meta Qwen Tencent
Own leaderboard Not tested since Aug 2026
GLM-5.3 61.4%
Lab claims 5 leaders

Sources: Official JobBench leaderboard and labs’ model cards

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 Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Muse Wonder Full Muse Spark 1.3 Claude Fable 5 Claude Opus 4.8 Claude Sonnet 4.6 Claude Sonnet 4.5 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Muse Wonder Full Muse Spark 1.3

Sources: Official JobBench leaderboard and labs’ model cards

8 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.

Meta
Moonshot AI
Qwen
Tencent
StepFun
Xiaomi Corp
Meituan
Thinking Machines
DateLaunch postScoreAs the post writes it
Nov 20, 2026 Inkling 2 Mini 49.6 JobBench
Nov 15, 2026 Qwen4-Max 68.5 Professional job tasks, JobBench (updated dataset and evaluation code)
Nov 5, 2026 Inkling 2 Small 70.5 JobBench
Nov 2, 2026 LongCat-2.5 65.4 JobBench
Oct 15, 2026 Muse Spark Open 61.8 JobBench
Oct 12, 2026 Muse Wonder 70.5 JobBench
Oct 12, 2026 Muse Wonder Full 75.3 JobBench
Sep 22, 2026 MiMo-V2.6 Pro 62 JobBench
Sep 21, 2026 MiMo-V2.6 Flash 61.2 JobBench
Sep 20, 2026 Step 5 Preview 59 JobBench
Sep 2, 2026 Qwen3.8 Max (0902) 64 Professional job tasks, JobBench
Sep 2, 2026 Muse Spark 1.3 64.9 JobBench (Professional tool use)
Aug 28, 2026 Tencent Hy4 preview 61.7 JobBench
Aug 26, 2026 Qwen3.8 Flash 55.7 Professional job tasks JobBench
Aug 14, 2026 Qwen 3.8 27B 33.4 Professional job tasks JobBench
Aug 3, 2026 Qwen 3.8 Max 53.4 JobBench
Jul 16, 2026 Kimi K3 54.3 JobBench
Jul 9, 2026 Muse Spark 1.1 54.7 JobBench (Professional tool use)

6 system cards report it, from 4 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
Nov 20, 2026 Inkling 2 Mini model card 49.6 JobBench Capabilities
Oct 12, 2026 Muse Wonder model card 70.5 JobBench mean reward, Muse Wonder (max) (capabilities table image, page 3) Capabilities
Oct 12, 2026 Muse Wonder Full model card 75.3 JobBench mean reward, Muse Wonder Full (max) (capabilities table image, page 3) Capabilities
Sep 22, 2026 MiMo-V2.6 Pro technical report 62 JobBench (Table 3) Capabilities
Jul 27, 2026 Kimi K3 technical report 54.3 JobBench, cited from the official leaderboard (Table 2) Capabilities
Jul 9, 2026 Muse Spark 1.1 system card 54.7 JobBench, OpenCode harness, Grok 4.3 judge (Figure 44) Capabilities

3 lab claims 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. 2 of them match, which usually means one side copied the other. Every benchmark.

Model Lab says Board says Gap
Qwen 3.8 Max 53.4% JobBench 52.7% Leaderboard, OpenCode +0.7 pts
Muse Spark 1.1 54.7% JobBench (Professional tool use) 54.7% Leaderboard, OpenCode Same
Kimi K3 54.3% JobBench 54.3% Leaderboard, OpenCode Same

More in Economic work

All 210 in Economic work
  • GDPval In use Real work tasks from 44 occupations with reference files. Experts blindly compare model deliverables with expert work. Metric is win-or-tie rate. Seed2.1 Pro 87.9%
  • GDPval-AA v2.1 In use Agents with shell and web access complete 220 GDPval gold tasks from 44 occupations. Blind pairwise comparisons give Elo ratings. Claude Opus 5.5 1,846
  • AA-Briefcase v1.1 In use Version 1.1 of Artificial Analysis's multi-hour office work evaluation, with Elo ratings fitted by Crowd-BT. Claude Opus 5.5 1,822
  • Agents' Last Exam v1.1 In use Agents complete 122 public long-horizon professional workflow tasks across 55 sub-industries. Deterministic graders check outcomes against hidden references. GPT-6.2 Astra 47.5%
  • CoWorkBench In use Agent completes long-horizon productivity tasks across computer science, finance, law and medicine. Qwen in-house scoring (method unpublished). Claude Fable 5 75.9%
  • Vending-Bench 2 In use Agent runs a simulated vending-machine business for one year starting with $500. Score is the final bank balance, averaged over runs. Gemini 4 Neon $16,843

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 (16 entries), launch posts (18 scores) and system cards (6 scores).
  • Pass Models scored 31 models we track.
  • Pass Newest score The newest is from November 2026, for Inkling 2 Mini.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Claude Fable 5 (#12).
  • Pass Headroom The best is 75.3%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on job-bench.github.io: 16 rows for models we track at the last read.
  • Pass Last read November 19th, 2026.
  • Pass Read worked The last read worked.
Consistency 2 pass
  • Pass Lab claims All 3 pairs agree within 10 points.
  • Pass Known issues No published audit found problems.

How we check