01.AILanguage model
Yi-34B
Yi-34B has no overall index value yet because there are too few independent benchmark results.
PopularityRank 431 of 785
34out of 100-2 vs. last month
based on search, usage and attention
Summary
Yi-34B is 01.AI’s earlier 34B base checkpoint. The official model card separates the base model from a separately instruction-tuned chat checkpoint. The base variant is intended as a versatile foundation for custom fine-tuning, while the chat version is adapted for conversation. Yi-34B is therefore a base model rather than a ready-made assistant.
Yi-34B has a 4,096-token context. Yi-1.5 34B updates the same family with additional pretraining and post-training, but it is a different checkpoint. Yi-34B also keeps base and chat weights separate.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 33 published benchmark results
Page 1 of 3
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
OpenMedLM study MedQA (4 options) (Show measurement, test conditions, and source)
- Source value
- 58.4
- Score
- 58.4
- Metric
- accuracy
- Unit
- %
- Benchmark version
- OpenMedLM study, Table 1
- Category
- medical-qa
- Direction
- Higher is better
- Sample
- 1273
- Source type
- independent-evaluator
- Evaluator
- OpenMedLM paper, Table 1 and Methods
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Exact Yi 34B open foundation model, zero-shot baseline (no OpenMedLM prompt engineering). Accuracy on the dataset subset specified in the paper. English test subset, 1,273 questions.
- Context
- The source publishes no confidence interval and does not pin an immutable Yi checkpoint revision.
OpenMedLM study MedMCQA (Show measurement, test conditions, and source)
- Source value
- 55.9
- Score
- 55.9
- Metric
- accuracy
- Unit
- %
- Benchmark version
- OpenMedLM study, Table 1
- Category
- medical-qa
- Direction
- Higher is better
- Sample
- 4183
- Source type
- independent-evaluator
- Evaluator
- OpenMedLM paper, Table 1 and Methods
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Exact Yi 34B open foundation model, zero-shot baseline (no OpenMedLM prompt engineering). Accuracy on the dataset subset specified in the paper. dev subset, 4,183 questions.
- Context
- The source publishes no confidence interval and does not pin an immutable Yi checkpoint revision.
OpenMedLM study PubMedQA (Show measurement, test conditions, and source)
- Source value
- 53.4
- Score
- 53.4
- Metric
- accuracy
- Unit
- %
- Benchmark version
- OpenMedLM study, Table 1
- Category
- medical-qa
- Direction
- Higher is better
- Sample
- 500
- Source type
- independent-evaluator
- Evaluator
- OpenMedLM paper, Table 1 and Methods
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Exact Yi 34B open foundation model, zero-shot baseline (no OpenMedLM prompt engineering). Accuracy on the dataset subset specified in the paper. reasoning-required test subset, 500 questions.
- Context
- The source publishes no confidence interval and does not pin an immutable Yi checkpoint revision.
OpenMedLM study MMLU Medical subset (Show measurement, test conditions, and source)
- Source value
- 72.6
- Score
- 72.6
- Metric
- accuracy
- Unit
- %
- Benchmark version
- OpenMedLM study, Table 1
- Category
- medical-qa
- Direction
- Higher is better
- Sample
- 1785
- Source type
- independent-evaluator
- Evaluator
- OpenMedLM paper, Table 1 and Methods
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Exact Yi 34B open foundation model, zero-shot baseline (no OpenMedLM prompt engineering). Accuracy on the dataset subset specified in the paper. nine medical/clinical topics, 1,785 unique questions.
- Context
- The source publishes no confidence interval and does not pin an immutable Yi checkpoint revision.
SpAMCQA Single-choice accuracy (Show measurement, test conditions, and source)
- Source value
- 79.38
- Score
- 79.38
- Metric
- Single-choice accuracy
- Unit
- %
- Benchmark version
- SpAMCQA, 2026 paper Table 4
- Category
- medical-qa
- Direction
- Higher is better
- Sample
- 194
- Source type
- independent-evaluator
- Evaluator
- SpAMCQA paper, Table 4
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Exact public Yi-34B base model, zero-shot before fine-tuning. SpAMCQA uses a zero-shot chain-of-thought strategy for the evaluation, with single-choice, multi-choice, and overall accuracy reported separately. Question denominator: 194 single-choice questions.
- Context
- No confidence interval or immutable weight revision is reported. The multi-choice score uses the paper’s partial-credit scoring rule.
SpAMCQA Multi-choice accuracy (Show measurement, test conditions, and source)
- Source value
- 49.29
- Score
- 49.29
- Metric
- Multi-choice accuracy
- Unit
- %
- Benchmark version
- SpAMCQA, 2026 paper Table 4
- Category
- medical-qa
- Direction
- Higher is better
- Sample
- 28
- Source type
- independent-evaluator
- Evaluator
- SpAMCQA paper, Table 4
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Exact public Yi-34B base model, zero-shot before fine-tuning. SpAMCQA uses a zero-shot chain-of-thought strategy for the evaluation, with single-choice, multi-choice, and overall accuracy reported separately. Question denominator: 28 multi-choice questions.
- Context
- No confidence interval or immutable weight revision is reported. The multi-choice score uses the paper’s partial-credit scoring rule.
SpAMCQA Overall accuracy (Show measurement, test conditions, and source)
- Source value
- 75.59
- Score
- 75.59
- Metric
- Overall accuracy
- Unit
- %
- Benchmark version
- SpAMCQA, 2026 paper Table 4
- Category
- medical-qa
- Direction
- Higher is better
- Sample
- 222
- Source type
- independent-evaluator
- Evaluator
- SpAMCQA paper, Table 4
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Exact public Yi-34B base model, zero-shot before fine-tuning. SpAMCQA uses a zero-shot chain-of-thought strategy for the evaluation, with single-choice, multi-choice, and overall accuracy reported separately. Question denominator: 222 total questions.
- Context
- No confidence interval or immutable weight revision is reported. The multi-choice score uses the paper’s partial-credit scoring rule.
Synthetic EHR generation EPS (Show measurement, test conditions, and source)
- Source value
- 96.8
- Score
- 96.8
- Metric
- Electronic health record performance score (EPS), the percentage of fully complete and accurate generated EHRs among all generated records.
- Unit
- %
- Benchmark version
- JMIR comparative synthetic-EHR study, Table 2
- Direction
- Higher is better
- Sample
- 20000
- Source type
- independent-evaluator
- Evaluator
- Evaluation and Bias Analysis of Large Language Models in Generating Synthetic Electronic Health Records: Comparative Study
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Independent evaluation. Exact paper model row: “Yi-34B” (34B parameters; publisher listed as 01.AI). The study used ten prompts per disease, 100 cases per prompt, across 20 diseases, for 20,000 synthetic EHRs per model. The paper reports using Python and study-wide hardware of four NVIDIA 3090 GPUs and two A800 GPUs, but does not identify the per-model inference route (local weights, API, or other), immutable weight revision, decoding settings, or exact model-run date.
- Context
- No confidence interval is reported. The 20,000 generated EHRs per model are the source-stated EPS denominator; the paper does not separately report valid-output counts. Table 1 lists Yi-34B MMLU as 79.4, whereas Table 2 lists 74.9; those background MMLU/C-Eval values are not imported as new measurements.
Synthetic EHR race-attribute EPS (Show measurement, test conditions, and source)
- Source value
- 96.84
- Score
- 96.84
- Metric
- Race-specific electronic health record performance score (EPS_race), the percentage of records containing the specified attribute and otherwise fully accurate.
- Unit
- %
- Benchmark version
- JMIR comparative synthetic-EHR study, Table 2
- Direction
- Higher is better
- Sample
- 20000
- Source type
- independent-evaluator
- Evaluator
- Evaluation and Bias Analysis of Large Language Models in Generating Synthetic Electronic Health Records: Comparative Study
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Independent evaluation. Exact paper model row: “Yi-34B” (34B parameters; publisher listed as 01.AI). The study used ten prompts per disease, 100 cases per prompt, across 20 diseases, for 20,000 synthetic EHRs per model. The paper reports using Python and study-wide hardware of four NVIDIA 3090 GPUs and two A800 GPUs, but does not identify the per-model inference route (local weights, API, or other), immutable weight revision, decoding settings, or exact model-run date.
- Context
- No confidence interval is reported. The 20,000 generated EHRs per model are the source-stated EPS denominator; the paper does not separately report valid-output counts. Table 1 lists Yi-34B MMLU as 79.4, whereas Table 2 lists 74.9; those background MMLU/C-Eval values are not imported as new measurements.
Synthetic EHR gender-attribute EPS (Show measurement, test conditions, and source)
- Source value
- 98.82
- Score
- 98.82
- Metric
- Gender-specific electronic health record performance score (EPS_gender), the percentage of records containing the specified attribute and otherwise fully accurate.
- Unit
- %
- Benchmark version
- JMIR comparative synthetic-EHR study, Table 2
- Direction
- Higher is better
- Sample
- 20000
- Source type
- independent-evaluator
- Evaluator
- Evaluation and Bias Analysis of Large Language Models in Generating Synthetic Electronic Health Records: Comparative Study
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Independent evaluation. Exact paper model row: “Yi-34B” (34B parameters; publisher listed as 01.AI). The study used ten prompts per disease, 100 cases per prompt, across 20 diseases, for 20,000 synthetic EHRs per model. The paper reports using Python and study-wide hardware of four NVIDIA 3090 GPUs and two A800 GPUs, but does not identify the per-model inference route (local weights, API, or other), immutable weight revision, decoding settings, or exact model-run date.
- Context
- No confidence interval is reported. The 20,000 generated EHRs per model are the source-stated EPS denominator; the paper does not separately report valid-output counts. Table 1 lists Yi-34B MMLU as 79.4, whereas Table 2 lists 74.9; those background MMLU/C-Eval values are not imported as new measurements.
MMLU-Pro (Show measurement, test conditions, and source)
- Source value
- 43.03
- Score
- 43.03
- Metric
- score
- Unit
- No unit provided
- Benchmark version
- TIGER-Lab MMLU-Pro leaderboard result
- Direction
- Higher is better
- Source type
- official-leaderboard
- Evaluator
- TIGER-Lab MMLU-Pro official leaderboard
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Official benchmark leaderboard. Exact 01-ai/Yi-34B model result linked by the model card to the MMLU-Pro leaderboard; prompt, shot count, quantization, and run date are not stated. Benchmark prompt, sampling/decoding settings, reasoning effort, quantization, and task run date are not stated in the cited result; none are inferred.
- Context
- The source presents a raw score without an explicit unit or scale; the published value is retained without percentage conversion.
H6 (Show measurement, test conditions, and source)
- Source value
- 69.42
- Score
- 69.42
- Metric
- six-task aggregate
- Unit
- No unit provided
- Benchmark version
- Open LLM Leaderboard snapshot in SOLAR technical report Table 2
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SOLAR technical report
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Yi-34B pretrained (not 34B-200K and not Chat); exact Table 2 row; raw score, unit not printed.
- Context
- The source does not provide a sample count or confidence interval for this result.
Page 1 of 3
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
The 100 most popular models of this category, 0 to 100, higher is more popular
20 of 101 models
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 7, 2026
Yi-34BMiMo-V2-Flash 35Tiny Aya Global 34GPT-4.1 nano 34Grok Build 0.1 34
Method changes in the timeline, a jump at these points is no market event:
- 1November 1, 2026: Value only with search or usage data, 30-day attention
What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.
Similar models
More models of the same kind as Yi-34B, chosen by model line, developer, size, price and release date.
- Yi-1.5 34B01.AI · same developer
- Granite Code 34BIBM · similar size
- Mixtral 8x7BMistral AI · similar size
- Qwen2.5-32B-InstructAlibaba · similar size
- Qwen 2.5 Coder 32BAlibaba · similar size
- InternLM2-Chat-20BShanghai AI Lab · similar size
- DeepSeek LLM 67BDeepSeek · similar size
- Qwen 3 32BAlibaba · similar size
All specifications
Architecture
Input and output
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
Frequently asked questions about Yi-34B
How large is the context window of Yi-34B?
4,096 tokens. That is how much text Yi-34B reads at once, about 10 A4 pages.
Page count as a rule of thumb from Artificial Analysis
GLM-5.1200K
DeepSeek-V3 0324164K
GLM-4.5131K
GPT OSS 120B131K
DeepSeek-V3.2128K
DeepSeek-R1128K
Yi-34BRank 103 of 1034K
Can I run Yi-34B locally?
Yes, the weights are open.
Stated by 01.AI
- Open weights
- yes
- License
- Apache 2.0
- Parameters
- 34B
- active per token
- n/a
Sources and data date
Show all 15 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Without vendor claims and historical or disputed results; those stay labelled in the single values.