Codesota · Reasoning · Multi-step Reasoning · BIG-Bench HardTasks/Reasoning/Multi-step Reasoning
Multi-step Reasoning · benchmark dataset · 2022 · EN

BIG-Bench Hard (BBH).

BIG-Bench Hard is a curated subset of 23 challenging tasks from BIG-Bench that require multi-step reasoning, where chain-of-thought prompting significantly helps performance. Tasks include algorithmic reasoning, logical deduction, causal judgment, and more. By 2024–2025, frontier models were approaching saturation (>90%) on BBH, prompting the creation of the harder BBEH variant.

Paper ↗Submit a result ↵
§ 01 · Leaderboard

Best published scores.

11 results indexed across 1 metric. Shaded row marks current SOTA; ties broken by submission date.

Primary
accuracy · higher is better
accuracy· primary
11 rows
#ModelOrgSubmittedPaper / codeaccuracy
01Claude 3.5 SonnetAPIAnthropicMar 2026llm-stats-bbh93.10
02Gemini 1.5 ProAPIGoogleMar 2026llm-stats-bbh89.20
03Qwen3-235B-A22BOpenAlibabaMay 2025Qwen3 Technical Report · code88.87
04Step-3.5-Flash Base—Feb 2026Step 3.5 Flash: Open Frontier-Level Intelligence with 11… · code88.20
05Gemma-3-27bOpenGoogleMar 2026llm-stats-bbh87.60
06Claude 3 OpusAPIAnthropicMar 2026llm-stats-bbh86.80
07Llama 3.1 405BOpenMetaMar 2026llm-stats-bbh85.90
08MiniCPM-o 4.5-Instruct—Apr 2026MiniCPM-o 4.5: Towards Real-Time Full-Duplex Omni-Modal … · code81.10
09Apertus-70B-Instruct—Sep 2025Apertus: Democratizing Open and Compliant LLMs for Globa… · code64.20
10Llama 2 70B (5-shot)—Jul 2023Llama 2: Open Foundation and Fine-Tuned Chat Models · code51.20
11SmoLM2 (1.7B)—Feb 2025SmolLM2: When Smol Goes Big -- Data-Centric Training of … · code32.20
Fig 2 · Rows sorted by score within each metric. Shaded row marks SOTA. Dates reflect model or paper release where available, otherwise the date Codesota accessed the source.
§ 03 · Progress

3 steps
of state of the art.

Each row below marks a model that broke the previous record on accuracy. Intermediate submissions are kept in the leaderboard above; only SOTA-setting entries are re-listed here.

Higher scores win. Each subsequent entry improved upon the previous best.

SOTA line · accuracy
  1. Jul 18, 2023Llama 2 70B (5-shot)51.20
  2. May 14, 2025Qwen3-235B-A22BAlibaba88.87
  3. Mar 28, 2026Claude 3.5 SonnetAnthropic93.10
Fig 3 · SOTA-setting models only. 3 entries span Jul 2023 → Mar 2026.
§ 04 · Literature

6 papers
tied to this benchmark.

Every paper below corresponds to at least one row in the leaderboard above. Click through for the arXiv preprint and, when available, the reference implementation.

§ 06 · Contribute

Have a score that beats
this table?

Submit a checkpoint and a reproduction script. We will run it, publish the score, and — if it takes the top — annotate the step on the progress chart with your name.

Submit a result ↵Read submission guide
What a submission needs
  • 01A public checkpoint or API endpoint
  • 02A reproduction script with frozen commit + seed
  • 03Declared evaluation environment (Python, deps)
  • 04One row per metric declared by this dataset
  • 05A contact so we can follow up on discrepancies
BIG-Bench Hard — Multi-step Reasoning | CodeSOTA