Shanghai AI LabLanguage model
InternLM3 8B-Instruct
InternLM3 8B-Instruct has no overall index value yet because there are too few independent benchmark results.
PopularityRank 506 of 785
31out of 100-1 vs. last month
based on search, usage and attention
Summary
InternLM3 8B-Instruct is an 8B model for general tasks and advanced reasoning. Its most useful interaction idea is switching between a normal response and a deep-thinking mode within the same checkpoint. That mode switch saves latency on simple requests and allows more deliberate work on harder ones without changing model IDs. The card also describes pretraining on 4 trillion high-quality tokens.
The checkpoint has a 32,768-token context and uses Apache 2.0. InternLM’s lower training-cost figure describes model development, not an independent runtime result.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 102 published benchmark results
Page 1 of 9
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.
Table 4, Computer Science, Chain-of-Thought, Uniqueness (Show measurement, test conditions, and source)
- Source value
- 94.81
- Score
- 94.81
- Metric
- Uniqueness rate
- Unit
- %
- Benchmark version
- KDD 2026 evaluation, Table 4
- Category
- reasoning
- Direction
- Higher is better
- Sample
- 1406
- Source type
- independent-evaluator
- Evaluator
- KDD 2026 academic evaluation paper
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. InternLM3-8B-Instruct; the paper reports 1,406 generated Computer Science questions for each of Chain-of-Thought and Free-Form Generation Prompt, and 438 generated K-12 Math questions for Free-Form Generation Prompt. Settings are temperature 0.9 and maximum generation length 500 tokens. Table 4 scores these generated questions; no confidence interval is reported.
- Context
- These are submetrics from one academic paper, not independent benchmark studies. Confidence intervals are not reported. Sample counts are included only where the source or its appendix gives the relevant metric denominator.
Table 4, Computer Science, Chain-of-Thought, Readability (Show measurement, test conditions, and source)
- Source value
- 99.86
- Score
- 99.86
- Metric
- Readability rate
- Unit
- %
- Benchmark version
- KDD 2026 evaluation, Table 4
- Category
- reasoning
- Direction
- Higher is better
- Sample
- 1406
- Source type
- independent-evaluator
- Evaluator
- KDD 2026 academic evaluation paper
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. InternLM3-8B-Instruct; the paper reports 1,406 generated Computer Science questions for each of Chain-of-Thought and Free-Form Generation Prompt, and 438 generated K-12 Math questions for Free-Form Generation Prompt. Settings are temperature 0.9 and maximum generation length 500 tokens. Table 4 scores these generated questions; no confidence interval is reported.
- Context
- These are submetrics from one academic paper, not independent benchmark studies. Confidence intervals are not reported. Sample counts are included only where the source or its appendix gives the relevant metric denominator.
Table 4, Computer Science, Chain-of-Thought, Answerability (Show measurement, test conditions, and source)
- Source value
- 99.79
- Score
- 99.79
- Metric
- Answerability rate
- Unit
- %
- Benchmark version
- KDD 2026 evaluation, Table 4
- Category
- reasoning
- Direction
- Higher is better
- Sample
- 1406
- Source type
- independent-evaluator
- Evaluator
- KDD 2026 academic evaluation paper
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. InternLM3-8B-Instruct; the paper reports 1,406 generated Computer Science questions for each of Chain-of-Thought and Free-Form Generation Prompt, and 438 generated K-12 Math questions for Free-Form Generation Prompt. Settings are temperature 0.9 and maximum generation length 500 tokens. Table 4 scores these generated questions; no confidence interval is reported.
- Context
- These are submetrics from one academic paper, not independent benchmark studies. Confidence intervals are not reported. Sample counts are included only where the source or its appendix gives the relevant metric denominator.
Table 4, Computer Science, Free-Form Generation Prompt, Uniqueness (Show measurement, test conditions, and source)
- Source value
- 98.94
- Score
- 98.94
- Metric
- Uniqueness rate
- Unit
- %
- Benchmark version
- KDD 2026 evaluation, Table 4
- Category
- reasoning
- Direction
- Higher is better
- Sample
- 1406
- Source type
- independent-evaluator
- Evaluator
- KDD 2026 academic evaluation paper
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. InternLM3-8B-Instruct; the paper reports 1,406 generated Computer Science questions for each of Chain-of-Thought and Free-Form Generation Prompt, and 438 generated K-12 Math questions for Free-Form Generation Prompt. Settings are temperature 0.9 and maximum generation length 500 tokens. Table 4 scores these generated questions; no confidence interval is reported.
- Context
- These are submetrics from one academic paper, not independent benchmark studies. Confidence intervals are not reported. Sample counts are included only where the source or its appendix gives the relevant metric denominator.
Table 4, Computer Science, Free-Form Generation Prompt, Readability (Show measurement, test conditions, and source)
- Source value
- 100
- Score
- 100
- Metric
- Readability rate
- Unit
- %
- Benchmark version
- KDD 2026 evaluation, Table 4
- Category
- reasoning
- Direction
- Higher is better
- Sample
- 1406
- Source type
- independent-evaluator
- Evaluator
- KDD 2026 academic evaluation paper
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. InternLM3-8B-Instruct; the paper reports 1,406 generated Computer Science questions for each of Chain-of-Thought and Free-Form Generation Prompt, and 438 generated K-12 Math questions for Free-Form Generation Prompt. Settings are temperature 0.9 and maximum generation length 500 tokens. Table 4 scores these generated questions; no confidence interval is reported.
- Context
- These are submetrics from one academic paper, not independent benchmark studies. Confidence intervals are not reported. Sample counts are included only where the source or its appendix gives the relevant metric denominator.
Table 4, Computer Science, Free-Form Generation Prompt, Answerability (Show measurement, test conditions, and source)
- Source value
- 99.93
- Score
- 99.93
- Metric
- Answerability rate
- Unit
- %
- Benchmark version
- KDD 2026 evaluation, Table 4
- Category
- reasoning
- Direction
- Higher is better
- Sample
- 1406
- Source type
- independent-evaluator
- Evaluator
- KDD 2026 academic evaluation paper
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. InternLM3-8B-Instruct; the paper reports 1,406 generated Computer Science questions for each of Chain-of-Thought and Free-Form Generation Prompt, and 438 generated K-12 Math questions for Free-Form Generation Prompt. Settings are temperature 0.9 and maximum generation length 500 tokens. Table 4 scores these generated questions; no confidence interval is reported.
- Context
- These are submetrics from one academic paper, not independent benchmark studies. Confidence intervals are not reported. Sample counts are included only where the source or its appendix gives the relevant metric denominator.
Table 4, K-12 Math, Free-Form Generation Prompt, Uniqueness (Show measurement, test conditions, and source)
- Source value
- 97.07
- Score
- 97.07
- Metric
- Uniqueness rate
- Unit
- %
- Benchmark version
- KDD 2026 evaluation, Table 4
- Category
- reasoning
- Direction
- Higher is better
- Sample
- 438
- Source type
- independent-evaluator
- Evaluator
- KDD 2026 academic evaluation paper
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. InternLM3-8B-Instruct; the paper reports 1,406 generated Computer Science questions for each of Chain-of-Thought and Free-Form Generation Prompt, and 438 generated K-12 Math questions for Free-Form Generation Prompt. Settings are temperature 0.9 and maximum generation length 500 tokens. Table 4 scores these generated questions; no confidence interval is reported.
- Context
- These are submetrics from one academic paper, not independent benchmark studies. Confidence intervals are not reported. Sample counts are included only where the source or its appendix gives the relevant metric denominator.
Table 4, K-12 Math, Free-Form Generation Prompt, Readability (Show measurement, test conditions, and source)
- Source value
- 100
- Score
- 100
- Metric
- Readability rate
- Unit
- %
- Benchmark version
- KDD 2026 evaluation, Table 4
- Category
- reasoning
- Direction
- Higher is better
- Sample
- 438
- Source type
- independent-evaluator
- Evaluator
- KDD 2026 academic evaluation paper
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. InternLM3-8B-Instruct; the paper reports 1,406 generated Computer Science questions for each of Chain-of-Thought and Free-Form Generation Prompt, and 438 generated K-12 Math questions for Free-Form Generation Prompt. Settings are temperature 0.9 and maximum generation length 500 tokens. Table 4 scores these generated questions; no confidence interval is reported.
- Context
- These are submetrics from one academic paper, not independent benchmark studies. Confidence intervals are not reported. Sample counts are included only where the source or its appendix gives the relevant metric denominator.
Table 4, K-12 Math, Free-Form Generation Prompt, Answerability (Show measurement, test conditions, and source)
- Source value
- 99.76
- Score
- 99.76
- Metric
- Answerability rate
- Unit
- %
- Benchmark version
- KDD 2026 evaluation, Table 4
- Category
- reasoning
- Direction
- Higher is better
- Sample
- 438
- Source type
- independent-evaluator
- Evaluator
- KDD 2026 academic evaluation paper
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. InternLM3-8B-Instruct; the paper reports 1,406 generated Computer Science questions for each of Chain-of-Thought and Free-Form Generation Prompt, and 438 generated K-12 Math questions for Free-Form Generation Prompt. Settings are temperature 0.9 and maximum generation length 500 tokens. Table 4 scores these generated questions; no confidence interval is reported.
- Context
- These are submetrics from one academic paper, not independent benchmark studies. Confidence intervals are not reported. Sample counts are included only where the source or its appendix gives the relevant metric denominator.
Table 5, Computer Science, Chain-of-Thought, Usability (Show measurement, test conditions, and source)
- Source value
- 91.33
- Score
- 91.33
- Metric
- Usability rate
- Unit
- %
- Benchmark version
- KDD 2026 evaluation, Table 5
- Category
- reasoning
- Direction
- Higher is better
- Sample
- 1406
- Source type
- independent-evaluator
- Evaluator
- KDD 2026 academic evaluation paper
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. InternLM3-8B-Instruct; Table 5 reports quality metrics. Usability uses the generated-question totals (1,406 per Computer Science prompt, 438 K-12 Math FGP). Category Consistency and Knowledge Coverage are evaluated only on valid and usable outputs; Appendix B gives the corresponding counts (1,332 CS CoT, 1,396 CS FGP, 397 K-12 Math FGP). Temperature 0.9; maximum generation length 500 tokens. No confidence interval is reported.
- Context
- These are submetrics from one academic paper, not independent benchmark studies. Confidence intervals are not reported. Sample counts are included only where the source or its appendix gives the relevant metric denominator.
Table 5, Computer Science, Chain-of-Thought, Category Consistency (Show measurement, test conditions, and source)
- Source value
- 32.53
- Score
- 32.53
- Metric
- Category Consistency rate
- Unit
- %
- Benchmark version
- KDD 2026 evaluation, Table 5
- Category
- reasoning
- Direction
- Higher is better
- Sample
- 1332
- Source type
- independent-evaluator
- Evaluator
- KDD 2026 academic evaluation paper
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. InternLM3-8B-Instruct; Table 5 reports quality metrics. Usability uses the generated-question totals (1,406 per Computer Science prompt, 438 K-12 Math FGP). Category Consistency and Knowledge Coverage are evaluated only on valid and usable outputs; Appendix B gives the corresponding counts (1,332 CS CoT, 1,396 CS FGP, 397 K-12 Math FGP). Temperature 0.9; maximum generation length 500 tokens. No confidence interval is reported.
- Context
- These are submetrics from one academic paper, not independent benchmark studies. Confidence intervals are not reported. Sample counts are included only where the source or its appendix gives the relevant metric denominator.
Table 5, Computer Science, Chain-of-Thought, Knowledge Coverage (Show measurement, test conditions, and source)
- Source value
- 76.26
- Score
- 76.26
- Metric
- Knowledge Coverage rate
- Unit
- %
- Benchmark version
- KDD 2026 evaluation, Table 5
- Category
- reasoning
- Direction
- Higher is better
- Sample
- 1332
- Source type
- independent-evaluator
- Evaluator
- KDD 2026 academic evaluation paper
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. InternLM3-8B-Instruct; Table 5 reports quality metrics. Usability uses the generated-question totals (1,406 per Computer Science prompt, 438 K-12 Math FGP). Category Consistency and Knowledge Coverage are evaluated only on valid and usable outputs; Appendix B gives the corresponding counts (1,332 CS CoT, 1,396 CS FGP, 397 K-12 Math FGP). Temperature 0.9; maximum generation length 500 tokens. No confidence interval is reported.
- Context
- These are submetrics from one academic paper, not independent benchmark studies. Confidence intervals are not reported. Sample counts are included only where the source or its appendix gives the relevant metric denominator.
Page 1 of 9
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
InternLM3 8B-InstructGPT-5 nano 31Qwen 3 Next 80B A3B Thinking 31Gemma 2 27B 31GPT-5.5 31
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 InternLM3 8B-Instruct, chosen by model line, developer, size, price and release date.
- InternLM2-Chat-20BShanghai AI Lab · same developer
- Aya Expanse 8BCohere · similar size
- ALLaM-2-7bSDAIA · similar size
- Qwen 2.5 7BAlibaba · similar size
- DeepSeek R1 0528 Qwen3 8BDeepSeek · similar size
- Command R7B ArabicCohere · similar size
- Command R7BCohere · similar size
- Aya Vision 8BCohere · 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 InternLM3 8B-Instruct
How large is the context window of InternLM3 8B-Instruct?
32,768 tokens. That is how much text InternLM3 8B-Instruct reads at once, about 50 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
InternLM3 8B-InstructRank 103 of 10333K
Can I run InternLM3 8B-Instruct locally?
Yes, the weights are open.
Stated by Shanghai AI Lab
- Open weights
- yes
- License
- Apache 2.0
- Parameters
- 8B
- active per token
- n/a
Sources and data date
Show all 13 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.