Grade School Math 8K (GSM8K)
We show this table for reference; we do not rank on it.
GSM8K tests multi-step arithmetic in grade-school word problems. OpenAI introduced the dataset in 2021; a high score on this fixed test is one piece of evidence about mathematical reasoning.
Benchmark score on GSM8K — October 29, 2026
We compile the GSM8K rows from provider self-reports. The table contains 9 models. OrcaCyber-Nano-1.0 has the highest published score at 97.5%, but the available coverage and evaluation setups do not establish a market leader. We do not use these results to rank models overall.
OrcaCyber-Nano-1.0
OrcaRouter
Ternary Bonsai 2 27B
Prism ML
Llama 3.1 70B
Meta
9 modelsMathematicsStaleSaturatedDisplay onlyUpdated October 29, 2026
Benchmark score table (9 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | OrcaCyber-Nano-1.0OrcaRouter | 97.5% | Not reported | Open |
| 2 | Ternary Bonsai 2 27BPrism ML | 96.7% | Not reported | Open |
| 3 | Llama 3.1 70BMeta | 95.1% | Not reported | Open |
| 4 | Celeris-1Celeris | 93.7% | Not reported | Closed |
| 5 | DeepSeek V4 Pro BaseDeepSeek | 92.6% | Not reported | Open |
| 6 | DeepSeek V4 Flash BaseDeepSeek | 90.8% | Not reported | Open |
| 7 | Phi-3.5-miniMicrosoft | 86.2% | Not reported | Open |
| 8 | Soofi S 30B-A3BSoofi Project | 86.1% | Not reported | Open |
| 9 | Llama 3.1 8BMeta | 84.5% | Not reported | Open |
How to read this leaderboard
GSM8K is a fixed dataset of human-written grade-school math word problems. It tests whether a model can carry a chain of elementary calculations through to the final answer.
Variant check: Match the prompt, number of examples, sampling rule, and answer extraction before comparing scores. GSM8K, translated MGSM, and MATH-500 answer different questions and retain separate tables.
A high GSM8K score is useful as a basic arithmetic check. When several rows approach the ceiling, use a harder math test and your own questions to distinguish them. We keep the published rows visible; a small set of self-reports does not establish a market leader.
Operator receipt: 9 sourced rows are currently displayable on this page; the highest published score among these 9 models is OrcaCyber-Nano-1.0 at 97.5%, which does not establish a market leader.
Provenance: The benchmark was introduced by Karl Cobbe and colleagues at OpenAI in 2021. A provider’s technical report supplies a result, not the authorship of GSM8K.
Among the reported GSM8K rows, OrcaCyber-Nano-1.0 is first at 97.5%. The third row is 2.4 points behind. The broader top-10 range is 13.0 points, so the table still separates the published systems.
9 models have been evaluated on GSM8K. The benchmark falls in the Mathematics category. GSM8K is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About GSM8K
Year
2021
Tasks
Grade-school math word problems
Format
Exact match
Difficulty
Grade-school math
GSM8K was introduced in Training Verifiers to Solve Math Word Problems. The public dataset contains human-written questions with worked solutions, generally requiring several arithmetic steps. Provider reports remain separate from the benchmark’s original authorship.
Freshness and provenance
Version
GSM8K 2021
Refresh cadence
Static
Staleness state
Stale
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does GSM8K measure?
GSM8K tests multi-step arithmetic in grade-school word problems. OpenAI introduced the dataset in 2021; a high score on this fixed test is one piece of evidence about mathematical reasoning.
Which model scores highest on GSM8K?
OrcaCyber-Nano-1.0 by OrcaRouter currently leads with a score of 97.5% on GSM8K.
How many models are evaluated on GSM8K?
9 AI models have published results on GSM8K in the BenchLM catalog.
Does a high GSM8K score identify the best math model?
A high GSM8K score supports a narrower claim: the evaluated setup solved many problems on this fixed grade-school test. Scores near the ceiling leave little room to separate models. Compare harder math tasks, check the prompting and sampling settings, and test questions that resemble your own workload.
Compare top models on GSM8K
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.