SDAIALanguage model
ALLaM-2-7b
ALLaM-2-7b has no overall index value yet because there are too few independent benchmark results.
PopularityRank 201 of 785
44out of 100+1 vs. last month
based on search, usage and attention
Summary
ALLaM-2-7b is an Arabic-English instruction checkpoint from SDAIA’s National Center for Artificial Intelligence. The official model card is named ALLaM-7B-Instruct-preview and describes a 7B model trained from scratch with a 4,096-token context. Pretraining combines 4 trillion English tokens with 1.2 trillion mixed English and Arabic tokens, followed by instruction and preference data.
ALLaM-2-7b is documented for Arabic-English language tooling rather than as a general 7B chatbot. The card shows an optional custom system prompt, although the model is optimized to work without a predefined prompt. The preview remains limited to 4K context and separates instruction and preference data in its post-training description.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 35 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.
MMLU English (Show measurement, test conditions, and source)
- Source value
- 63.65
- Score
- 63.65
- Metric
- Accuracy
- Unit
- %
- Benchmark version
- Groq documentation for API route allam-2-7b, checked 2026-10-17; benchmark revision and model serving revision not stated
- Category
- knowledge
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Groq model documentation: ALLaM-2-7B, route allam-2-7b
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. The route page labels the model ALLaM-2-7B and reports MMLU English (0-shot). Groq states that its served implementation uses TruePoint Numerics. No upstream weight/checkpoint mapping, serving revision, evaluation date, sample count, or other run settings are published.
- Context
- This is a dated observation for Groq's named API route, not a pinned checkpoint or independent replication. The upstream checkpoint bridge and original evaluation protocol are undisclosed.
Arabic MMLU (Show measurement, test conditions, and source)
- Source value
- 69.15
- Score
- 69.15
- Metric
- Accuracy
- Unit
- %
- Benchmark version
- Groq documentation for API route allam-2-7b, checked 2026-10-17; benchmark revision and model serving revision not stated
- Category
- arabic-knowledge
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Groq model documentation: ALLaM-2-7B, route allam-2-7b
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. The route page labels the model ALLaM-2-7B and reports Arabic MMLU (0-shot). Groq states that its served implementation uses TruePoint Numerics. No upstream weight/checkpoint mapping, serving revision, evaluation date, sample count, or other run settings are published.
- Context
- This is a dated observation for Groq's named API route, not a pinned checkpoint or independent replication. The upstream checkpoint bridge and original evaluation protocol are undisclosed.
ETEC Arabic (Show measurement, test conditions, and source)
- Source value
- 67
- Score
- 67
- Metric
- Accuracy
- Unit
- %
- Benchmark version
- Groq documentation for API route allam-2-7b, checked 2026-10-17; benchmark revision and model serving revision not stated
- Category
- arabic-knowledge
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Groq model documentation: ALLaM-2-7B, route allam-2-7b
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. The route page labels the model ALLaM-2-7B and reports ETEC Arabic (0-shot). Groq states that its served implementation uses TruePoint Numerics. No upstream weight/checkpoint mapping, serving revision, evaluation date, sample count, or other run settings are published.
- Context
- This is a dated observation for Groq's named API route, not a pinned checkpoint or independent replication. The upstream checkpoint bridge and original evaluation protocol are undisclosed.
IEN-MCQ (Show measurement, test conditions, and source)
- Source value
- 90.8
- Score
- 90.8
- Metric
- Accuracy
- Unit
- %
- Benchmark version
- Groq documentation for API route allam-2-7b, checked 2026-10-17; benchmark revision and model serving revision not stated
- Category
- arabic-knowledge
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Groq model documentation: ALLaM-2-7B, route allam-2-7b
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. The route page labels the model ALLaM-2-7B and reports IEN-MCQ accuracy. The source does not state shot count. Groq states that its served implementation uses TruePoint Numerics. No upstream weight/checkpoint mapping, serving revision, evaluation date, sample count, or other run settings are published.
- Context
- This is a dated observation for Groq's named API route, not a pinned checkpoint or independent replication. The upstream checkpoint bridge and original evaluation protocol are undisclosed.
MT-bench Arabic average (Show measurement, test conditions, and source)
- Source value
- 6.6
- Score
- 6.6
- Metric
- Average score on the published 0-10 scale
- Unit
- points
- Benchmark version
- MT-bench; revision, judge, prompt setup and evaluation date not stated
- Category
- arabic-dialogue
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Groq model documentation: ALLaM-2-7B, route allam-2-7b
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. The route page labels the model ALLaM-2-7B and reports MT-bench Arabic Average 6.6/10. Groq states TruePoint Numerics for its served implementation. No upstream weight/checkpoint mapping, serving revision, evaluation date, sample count, or judge setup are published.
- Context
- This is a dated observation for Groq's named API route, not a pinned checkpoint or independent replication. The upstream checkpoint bridge and MT-bench evaluation protocol are undisclosed.
MT-bench English average (Show measurement, test conditions, and source)
- Source value
- 7.14
- Score
- 7.14
- Metric
- Average score on the published 0-10 scale
- Unit
- points
- Benchmark version
- MT-bench; revision, judge, prompt setup and evaluation date not stated
- Category
- dialogue
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Groq model documentation: ALLaM-2-7B, route allam-2-7b
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. The route page labels the model ALLaM-2-7B and reports MT-bench English Average 7.14/10. Groq states TruePoint Numerics for its served implementation. No upstream weight/checkpoint mapping, serving revision, evaluation date, sample count, or judge setup are published.
- Context
- This is a dated observation for Groq's named API route, not a pinned checkpoint or independent replication. The upstream checkpoint bridge and MT-bench evaluation protocol are undisclosed.
MizanQA: PM(1) (Show measurement, test conditions, and source)
- Source value
- 26.88
- Score
- 26.88
- Metric
- PM(1)
- Unit
- No unit provided
- Category
- knowledge
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- EACL Anthology, MizanQA: Benchmarking Large Language Models on Moroccan Legal Question Answering
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Independent evaluation. The exact table label is Allam-2 (7b). The study evaluates models zero-shot in native Arabic script with a fixed prompt, temperature 1, no tools or retrieval, and per-option confidence output; malformed responses are re-prompted and remaining failures are counted incorrect. The table does not label the score scale. No N or CI is stated for this result.
- Context
- The paper describes MizanQA as containing 1,776 questions, but does not bind that total as the exact N for this model result. Table values are retained verbatim with unknown units; no conversion to percent is made.
MizanQA: PM(0.5) (Show measurement, test conditions, and source)
- Source value
- 34.04
- Score
- 34.04
- Metric
- PM(0.5)
- Unit
- No unit provided
- Category
- knowledge
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- EACL Anthology, MizanQA: Benchmarking Large Language Models on Moroccan Legal Question Answering
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Independent evaluation. The exact table label is Allam-2 (7b). The study evaluates models zero-shot in native Arabic script with a fixed prompt, temperature 1, no tools or retrieval, and per-option confidence output; malformed responses are re-prompted and remaining failures are counted incorrect. The table does not label the score scale. No N or CI is stated for this result.
- Context
- The paper describes MizanQA as containing 1,776 questions, but does not bind that total as the exact N for this model result. Table values are retained verbatim with unknown units; no conversion to percent is made.
MizanQA: F1(1) (Show measurement, test conditions, and source)
- Source value
- 43.07
- Score
- 43.07
- Metric
- F1(1)
- Unit
- No unit provided
- Category
- knowledge
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- EACL Anthology, MizanQA: Benchmarking Large Language Models on Moroccan Legal Question Answering
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Independent evaluation. The exact table label is Allam-2 (7b). The study evaluates models zero-shot in native Arabic script with a fixed prompt, temperature 1, no tools or retrieval, and per-option confidence output; malformed responses are re-prompted and remaining failures are counted incorrect. The table does not label the score scale. No N or CI is stated for this result.
- Context
- The paper describes MizanQA as containing 1,776 questions, but does not bind that total as the exact N for this model result. Table values are retained verbatim with unknown units; no conversion to percent is made.
MizanQA: F1(2) (Show measurement, test conditions, and source)
- Source value
- 39.93
- Score
- 39.93
- Metric
- F1(2)
- Unit
- No unit provided
- Category
- knowledge
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- EACL Anthology, MizanQA: Benchmarking Large Language Models on Moroccan Legal Question Answering
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Independent evaluation. The exact table label is Allam-2 (7b). The study evaluates models zero-shot in native Arabic script with a fixed prompt, temperature 1, no tools or retrieval, and per-option confidence output; malformed responses are re-prompted and remaining failures are counted incorrect. The table does not label the score scale. No N or CI is stated for this result.
- Context
- The paper describes MizanQA as containing 1,776 questions, but does not bind that total as the exact N for this model result. Table values are retained verbatim with unknown units; no conversion to percent is made.
MizanQA: ACC (Show measurement, test conditions, and source)
- Source value
- 15.32
- Score
- 15.32
- Metric
- ACC
- Unit
- No unit provided
- Category
- knowledge
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- EACL Anthology, MizanQA: Benchmarking Large Language Models on Moroccan Legal Question Answering
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Independent evaluation. The exact table label is Allam-2 (7b). The study evaluates models zero-shot in native Arabic script with a fixed prompt, temperature 1, no tools or retrieval, and per-option confidence output; malformed responses are re-prompted and remaining failures are counted incorrect. The table does not label the score scale. No N or CI is stated for this result.
- Context
- The paper describes MizanQA as containing 1,776 questions, but does not bind that total as the exact N for this model result. Table values are retained verbatim with unknown units; no conversion to percent is made.
MizanQA: ECEopt (Show measurement, test conditions, and source)
- Source value
- 28.42
- Score
- 28.42
- Metric
- ECEopt
- Unit
- No unit provided
- Category
- knowledge
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- EACL Anthology, MizanQA: Benchmarking Large Language Models on Moroccan Legal Question Answering
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Independent evaluation. The exact table label is Allam-2 (7b). The study evaluates models zero-shot in native Arabic script with a fixed prompt, temperature 1, no tools or retrieval, and per-option confidence output; malformed responses are re-prompted and remaining failures are counted incorrect. ECEopt is option-level Expected Calibration Error. The table does not label the score scale. No N or CI is stated for this result.
- Context
- The paper describes MizanQA as containing 1,776 questions, but does not bind that total as the exact N for this model result. Table values are retained verbatim with unknown units; no conversion to percent is made. ECE is an error measure, so lower is better.
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
ALLaM-2-7bGranite 4.1 8B 44GLM-5.3-Flash 44DeepSeek-R1-Distill-Llama-70B 44Gemini 3.1 Flash-Lite 44
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 ALLaM-2-7b, chosen by model line, developer, size, price and release date.
- Command R7BCohere · similar size
- Falcon 3 7BTII · similar size
- Qwen 2.5 7BAlibaba · similar size
- InternLM3 8B-InstructShanghai AI Lab · similar size
- Phi-4-multimodal-instructMicrosoft · similar size
- Command R7B ArabicCohere · similar size
- Gemma 3n E2B ITGoogle · similar size
- Mistral 7B v0.3Mistral AI · 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 ALLaM-2-7b
How large is the context window of ALLaM-2-7b?
4,096 tokens. That is how much text ALLaM-2-7b 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
ALLaM-2-7bRank 103 of 1034K
Can I run ALLaM-2-7b locally?
Yes, the weights are open.
Stated by SDAIA
- Open weights
- yes
- License
- Apache 2.0
- Parameters
- 7B
- active per token
- n/a
Sources and data date
Show all 14 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. Retrieved October 27, 2026. |
| Output | Documented modalities. Retrieved October 27, 2026. |
| Research status | Research date October 15, 2026. 6 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not published | score-units-and-harness Note The model card states the shot counts for these rows but not the score units, prompt details, evaluation dates, or aggregation protocol. |
| Unresolved | preview-checkpoint-bridge Note The preview card's results are not assigned to ALLaM-2-7B without an exact-checkpoint bridge. Separately sourced provider-route results do not prove identical upstream weights. |
| Unresolved | upstream-checkpoint-bridge-and-evaluation-protocol Note Groq's exact allam-2-7b route page names ALLaM-2-7B and reports the route-specific values, but does not disclose a serving revision, upstream checkpoint/weight mapping, original evaluation date, or full protocol. The existing ALLaM-7B-Instruct-preview benchmark rows do not establish this route's upstream checkpoint and must remain separate. |
| Additional source | Hugging Face · ALLaM 7B instruction model card · Editorial description reviewed October 2, 2026 |
| Additional source | Groq model documentation: ALLaM-2-7B, route allam-2-7b (retrieved October 17, 2026) |
| Additional source | EACL Anthology, MizanQA: Benchmarking Large Language Models on Moroccan Legal Question Answering (retrieved October 18, 2026) |
| Additional source | Mendacium PSYOP-Bench Track A (retrieved October 20, 2026) |
| Additional source | Exploring the Expressive Power of Large Language Models in Neuro-Fuzzy System Explainability: A Study on EEG-Based Seizure Detection (retrieved October 21, 2026) |
| Additional source | Model metadata (retrieved October 18, 2026) |
| Additional source | ALLaM research paper · Editorial description reviewed October 2, 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 |