Shanghai AI LabLanguage model
InternLM2-Chat-20B
InternLM2-Chat-20B has no overall index value yet because there are too few independent benchmark results.
22out of 100-8 vs. last month
based on search, usage and attention
Summary
InternLM2-Chat-20B is the dialogue-focused 20B checkpoint in the InternLM2 family. Its official card highlights improvements in reasoning, mathematics, code, instruction following, and creative work. It also documents a code interpreter, data analysis, and stronger tool use with multi-step tool calls.
For long inputs, the card documents an inference path of up to 200,000 tokens through LMDeploy. That window belongs to the LMDeploy configuration, while the 20B weight set still determines local resource needs. The chat template and backend are therefore part of the same InternLM2 integration.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 48 published benchmark results
Page 1 of 4
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.
IFEval (Show measurement, test conditions, and source)
- Source value
- 42.2
- Score
- 42.2
- Metric
- 0-100 accuracy
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- InternLM2 Technical Report, arXiv 2403.17297v1, Table 17
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. Exact Table 17 value on its stated 0-100 accuracy scale. Prompt templates are in Appendix A.2; run settings are unstated in the table.
ProteinLMBench, Correct Rate (Show measurement, test conditions, and source)
- Source value
- 51.38
- Score
- 51.38
- Metric
- Correct Rate
- Unit
- %
- Benchmark version
- proteinlmbench-paper-v2-full-official
- Category
- science
- Direction
- Higher is better
- Sample
- 944
- Source type
- independent-evaluator
- Evaluator
- TourSynbio paper, arXiv:2408.15299
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. TourSynbio's external evaluation of InternLM2-Chat-20B on ProteinLMBench; zero-shot, single-turn, public prompt, no browser, internet, databases, code execution, or external tools; prompt asks the model to think step by step; 20 generated tokens per question; temperature 0.1; one repeat; exact first-integer option match. The paper states 944 questions. It does not report a confidence interval, immutable checkpoint digest, or serving route.
- Context
- The BioBench Atlas audit finds that the current 944-row snapshot contains 871 six-choice questions, although the paper describes all 944 as six-choice. Preserve the paper's N=944 and disclose this mismatch; do not reinterpret the score.
MMLU (Show measurement, test conditions, and source)
- Source value
- 66.5
- Score
- 66.5
- Metric
- score
- Unit
- No unit provided
- Benchmark version
- InternLM2 OpenCompass table
- Category
- knowledge
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- InternLM2-Chat-20B official model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact InternLM2-Chat-20B model row/widget in InternLM2-Chat-20B official model card. 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.
AGIEval (Show measurement, test conditions, and source)
- Source value
- 50.3
- Score
- 50.3
- Metric
- score
- Unit
- No unit provided
- Benchmark version
- InternLM2 OpenCompass table
- Category
- knowledge
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- InternLM2-Chat-20B official model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact InternLM2-Chat-20B model row/widget in InternLM2-Chat-20B official model card. 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.
BBH (Show measurement, test conditions, and source)
- Source value
- 68.3
- Score
- 68.3
- Metric
- score
- Unit
- No unit provided
- Benchmark version
- InternLM2 OpenCompass table
- Category
- reasoning
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- InternLM2-Chat-20B official model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact InternLM2-Chat-20B model row/widget in InternLM2-Chat-20B official model card. 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.
GSM8K (Show measurement, test conditions, and source)
- Source value
- 79.6
- Score
- 79.6
- Metric
- score
- Unit
- No unit provided
- Benchmark version
- InternLM2 OpenCompass table
- Category
- math
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- InternLM2-Chat-20B official model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact InternLM2-Chat-20B model row/widget in InternLM2-Chat-20B official model card. 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.
MATH (Show measurement, test conditions, and source)
- Source value
- 31.9
- Score
- 31.9
- Metric
- score
- Unit
- No unit provided
- Benchmark version
- InternLM2 OpenCompass table
- Category
- math
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- InternLM2-Chat-20B official model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact InternLM2-Chat-20B model row/widget in InternLM2-Chat-20B official model card. 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.
HumanEval (Show measurement, test conditions, and source)
- Source value
- 67.1
- Score
- 67.1
- Metric
- score
- Unit
- No unit provided
- Benchmark version
- InternLM2 OpenCompass table
- Category
- coding
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- InternLM2-Chat-20B official model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact InternLM2-Chat-20B model row/widget in InternLM2-Chat-20B official model card. 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.
MBPP sanitized (Show measurement, test conditions, and source)
- Source value
- 65.8
- Score
- 65.8
- Metric
- score
- Unit
- No unit provided
- Benchmark version
- InternLM2 OpenCompass table
- Category
- coding
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- InternLM2-Chat-20B official model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact InternLM2-Chat-20B model row/widget in InternLM2-Chat-20B official model card. 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.
CMMLU (Show measurement, test conditions, and source)
- Source value
- 65.1
- Score
- 65.1
- Metric
- score
- Unit
- No unit provided
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- InternLM2 Technical Report, arXiv 2403.17297v1, Table 6
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. Exact InternLM2-Chat-20B row in Table 6, 5-shot. Sample count and decoding settings are not stated.
C-Eval (Show measurement, test conditions, and source)
- Source value
- 63
- Score
- 63
- Metric
- score
- Unit
- No unit provided
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- InternLM2 Technical Report, arXiv 2403.17297v1, Table 6
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. Exact InternLM2-Chat-20B row in Table 6, 5-shot. Sample count and decoding settings are not stated.
GAOKAO (Show measurement, test conditions, and source)
- Source value
- 58.3
- Score
- 58.3
- Metric
- score
- Unit
- No unit provided
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- InternLM2 Technical Report, arXiv 2403.17297v1, Table 6
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. Exact InternLM2-Chat-20B row in Table 6, 0-shot. Sample count and decoding settings are not stated.
Page 1 of 4
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
InternLM2-Chat-20BMiniMax M2.5 22Grok 4.3 22Pixtral Large (latest) 22Command A Reasoning 21
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 InternLM2-Chat-20B, chosen by model line, developer, size, price and release date.
- InternLM3 8B-InstructShanghai AI Lab · same developer
- StarCoder2 15BBigCode · similar size
- Codestral-22B-v0.1Mistral AI · similar size
- Codestral (latest)Mistral AI · similar size
- Phi-3 Medium (14B)Microsoft · similar size
- Yi-34B01.AI · similar size
- SOLAR 10.7B Instruct v1.0Upstage · similar size
- Phi-4Microsoft · 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 InternLM2-Chat-20B
Can I run InternLM2-Chat-20B locally?
Yes, the weights are open.
Stated by Shanghai AI Lab
- Open weights
- yes
- License
- Apache 2.0
- Parameters
- 20B
- active per token
- n/a
Sources and data date
Show all 10 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.
| Type | Evidence and data date |
|---|---|
| Input | Documented modalities. |
| Output | Documented modalities. |
| Research status | Research date October 15, 2026. 4 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not published | benchmark-task-date-and-evaluation-settings Note The exact Chat-20B column is attributed to OpenCompass but the visible card does not expose dataset revisions, prompts, scoring units, decoding settings, quantization, or run dates. |
| Not published | sample-count-and-uncertainty Note The tables do not give scored sample counts or confidence intervals for these model results. Dataset corpus sizes are not treated as denominators. |
| Unresolved | conflicting-model-card-scores Note For MATH, the report says 32.4 and the exact model card 31.9. For HumanEval, the report says 67.7 and the card 67.1. Causes and evaluation settings are unresolved. |
| Unresolved | table-8-shot-settings Note The table header and prose give different shot counts for TriviaQA, NaturalQuestions, C3, and FLORES. Scores are retained without silently choosing one setup. |
| Not published | needle-in-haystack-numeric-results Note The report presents the Needle-in-the-Haystack curve graphically without tabulated model values. No scores were estimated from the figure. |
| Additional source | InternLM2 Technical Report, arXiv 2403.17297v1, Table 6 (retrieved October 17, 2026) |
| Additional source | TourSynbio paper, arXiv:2408.15299 (retrieved October 17, 2026) |
| Additional source | MedOdyssey authors, arXiv:2406.15019, Table 2 (retrieved October 19, 2026) |
| Additional source | OpenRouter: Popularity index (derived), CC BY 4.0 |
| Additional source | Vercel AI Gateway: Popularity index (derived), CC BY 4.0 |
| Additional source | Ollama: cumulative downloads for the popularity index |
| Additional source | Replicate: cumulative runs for the popularity index |
| Additional source | Google Trending Now: limited feed selection for Germany and the US |
| Additional source | Creative Commons Attribution 4.0 |