| --- |
| library_name: transformers |
| license: apache-2.0 |
| pipeline_tag: text-generation |
| tags: |
| - moe |
| - vlm |
| - vision |
| - agentic |
| - code |
| - tool-use |
| --- |
| |
| # Sage-A1.5 |
|
|
| <div style="display: flex; flex-direction: column; align-items: center; line-height: 1.2;"> |
| <div style="display: flex; justify-content: center; align-items: center; gap: 10px; height: 30px;"> |
| <span style="font-size: 16px;" role="img" aria-label="Homepage">🏠</span> |
| <a href="https://alpha-innovator.github.io/Sage-A1.5/"><b>Homepage</b></a> |
| <span style="color: #ccc;">|</span> |
| <img src="./figures/24px.svg" width="16" height="16" alt="Technical Report" style="filter: invert(0.5);"> |
| <a href="https://arxiv.org/abs/2610.28578"><b>Technical Report</b></a> |
| <span style="color: #ccc;">|</span> |
| <span style="font-size: 16px;" role="img" aria-label="Free Trial">🎁</span> |
| <a href="https://discovery.intern-ai.org.cn/token-plan/home?tabIndex=0&lang=en-US"><b>Free Trial</b></a> |
| </div> |
| |
| <div style="display: flex; justify-content: center; align-items: center; gap: 10px; height: 30px; margin-top: 2px;"> |
| <img src="./figures/hf-logo.svg" width="16" height="16" alt="Hugging Face"> |
| <a href="https://huggingface.co/InternScience/Sage-A1.5"><b>Hugging Face</b></a> |
| <span style="color: #ccc;">|</span> |
| <img src="./figures/github-logo.svg" width="16" height="16" alt="GitHub"> |
| <a href="https://github.com/InternScience/Agents-A1"><b>Github</b></a> |
| <span style="color: #ccc;">|</span> |
| <img src="./figures/modelscope-logo.svg" width="16" height="16" alt="Model Scope"> |
| <a href="https://modelscope.cn/models/InternScience/Sage-A1.5"><b>ModelScope</b></a> |
| </div> |
| </div> |
| |
|
|
| > [!Note] |
| > This repository contains model weights and configuration files for Sage-A1.5 in the Hugging Face Transformers format. |
| > |
| > These artifacts are compatible with Hugging Face Transformers, vLLM, SGLang, etc. |
|
|
| --- |
|
|
| **Sage‑A1.5** is a 35B Mixture‑of‑Experts agentic model based on **[Qwen3.6](https://huggingface.co/Qwen/Qwen3.6-35B-A3B)** and **[Agents-A1](https://huggingface.co/InternScience/Agents-A1)** from Shanghai AI Laboratory, built to advance long-horizon scientific problem solving through adaptive capability composition. Building upon **A1**, **Sage‑A1.5** learns not only to solve individual tasks, but also to organize work and revise its approach during execution, highlighting three key capabilities: **Agentic Orchestration, Verification, and Open Scientific Discovery**. |
|
|
| First, agentic orchestration extends single-agent execution through task delegation and parallel collaboration among subagents with isolated contexts, |
| enabling complex tasks to be decomposed and addressed concurrently while reducing interference across concurrent subagents. |
| Second, agentic verification strengthens long-horizon reliability through progress checking, evidence validation, and reflective strategy revision, |
| allowing **Sage‑A1.5** to recognize unproductive search paths and pursue alternative approaches. |
| Third, open scientific discovery extends agentic capabilities beyond answering predefined questions to conducting exploratory research, |
| connecting scientific inquiry with interactive analysis and inspectable research artifacts. |
| We hope **Sage‑A1.5** provides the community with an accessible foundation for open scientific discovery, |
| offering a practical path from long-horizon task execution to collaborative, verifiable, and inspectable scientific exploration. |
|
|
|  |
|
|
| ## Highlights |
| - **Agentic Orchestration**: Orchestrates parallel investigations across subagents with independent contexts, enabling them to challenge assumptions, refine conclusions through evidence, and synthesize their findings into comprehensive results. |
| - **Verification Mode**: An opt-in self-review capability that lives inside the model's own chain of thought — the model pauses mid-trajectory to review what it has established and what has failed, then acts on a revised direction in the same turn, with no external critic. |
| - **Agentic Reasoning**: Sage-A1.5 excels at decomposing complex tasks into executable sub-steps, planning ahead, and adapting its strategy based on intermediate results. |
| - **Tool Use**: Natively supports function calling and tool integration, enabling seamless interaction with APIs, code interpreters, search engines, and other external tools. |
| - **Scientific and Professional Reasoning**: Handles tool-integrated scientific reasoning and professional knowledge question answering. |
| - **Instruction Following**: Precisely follows detailed, multi-constraint instructions across diverse domains. |
|
|
| We welcome developers and enterprises to integrate and try Sage-A1.5 and share their feedback. |
|
|
| ## Performance |
|
|
| We evaluate Sage-A1.5 in real-world agentic and research-oriented workflows across six |
| directions — long-horizon search, engineering tasks, instruction following, scientific research, |
| general agentic tasks, and scientific agentic tasks. Despite operating in the ~35B model class, |
| Sage-A1.5 delivers highly competitive performance against frontier-scale systems such as |
| Kimi-K3, GLM-5.3, and GPT-5.6-sol. It achieves overall SOTA on XBench-DS-2510 (92.0), |
| IFBench (81.6), PinchBench (90.0), FrontierScience-Olympiad (81.0), FrontierScience-Research (48.3), |
| CMT (80.0), and OmniDocBench (91.5), and reaches 95.1 on GAIA — higher than every |
| larger-scale model we compare against. Among comparable ~35B models it ranks first on a broad |
| range of tasks, including BrowseComp (81.7), SWE-bench Pro (61.2), MLE-Lite (65.2), |
| SciCode (44.3, tied with Agents-A1), AA-LCR (81.0), HLE w/ tools (55.1), |
| VitaBench (45.4), and ChemCOTBench (66.5). |
| Compared with A1, it improves substantially on engineering and agentic workloads |
| (SWE-bench Pro +17.6, SWE-bench Multilingual +14.0, PinchBench +13.1, MLE-Lite +21.3). |
| These results show that Sage-A1.5 combines strong long-horizon search ability, robust |
| scientific reasoning, and reliable instruction following, establishing it as a highly capable and |
| efficient agentic model that narrows the gap with much larger frontier models. |
|
|
|
|
| <p> |
| 🥇 Overall SOTA |
| 🟢 Best Among Comparable Models (~35B) |
| </p> |
|
|
| <table> |
| <thead> |
| <tr> |
| <th rowspan="2" align="left">Benchmark</th> |
| <th colspan="3" align="center" style="text-align:center;"> |
| 📏 Comparable Models (~35B) |
| </th> |
| |
| <th colspan="3" align="center" style="text-align:center;"> |
| 🚀 Larger-scale Models |
| </th> |
| </tr> |
| |
| <tr> |
| <th align="center">Sage-A1.5</th> |
| <th align="center">Agents-A1</th> |
| <th align="center">Ornith-1.5</th> |
| |
| <th align="center">Kimi-K3</th> |
| <th align="center">GLM-5.3</th> |
| <th align="center">GPT-5.6-sol</th> |
|
|
| </tr> |
| </thead> |
|
|
| <tbody> |
|
|
| <tr> |
| <td colspan="8" align="left"><b>🔍 Long-horizon Search</b></td> |
| </tr> |
| |
| <tr> |
| <td align="left">GAIA</td> |
| <td align="center">95.1</td> |
| <td align="center">96.0 🥇</td> |
| <td align="center">77.6</td> |
| <td align="center">92.2</td> |
| <td align="center">92.2</td> |
| <td align="center">88.3</td> |
| </tr> |
|
|
| <tr> |
| <td align="left">BrowseComp</td> |
| <td align="center">81.7 w/o Agentic Orchestration<br>82.8 🟢 w/ Agentic Orchestration</td> |
| <td align="center">75.5</td> |
| <td align="center">67.6</td> |
| <td align="center">91.2 🥇</td> |
| <td align="center">76.2</td> |
| <td align="center">90.4</td> |
| </tr> |
|
|
| <tr> |
| <td align="left">DeepResearch-Bench2</td> |
| <td align="center">43.7 w/o Agentic Orchestration<br>47.5 w/ Agentic Orchestration</td> |
| <td align="center">37.5</td> |
| <td align="center">48.6 🟢</td> |
| <td align="center">52.4</td> |
| <td align="center">55.7</td> |
| <td align="center">57.9 🥇</td> |
| </tr> |
|
|
| <tr> |
| <td align="left">XBench-DS-2510</td> |
| <td align="center">92.0 🥇</td> |
| <td align="center">86.0</td> |
| <td align="center">80.0</td> |
| <td align="center">85.0</td> |
| <td align="center">84.0</td> |
| <td align="center">90.0</td> |
| </tr> |
|
|
|
|
|
|
| <tr> |
| <td colspan="8" align="left"><b>⚙️ Engineering Tasks</b></td> |
| </tr> |
| |
| <tr> |
| <td align="left">MLE-Lite</td> |
| <td align="center">65.2 🟢</td> |
| <td align="center">43.9</td> |
| <td align="center">39.4</td> |
| <td align="center">72.7</td> |
| <td align="center">72.7</td> |
| <td align="center">77.3 🥇</td> |
| </tr> |
|
|
| <tr> |
| <td align="left">SciCode</td> |
| <td align="center">44.3 🟢</td> |
| <td align="center">44.3</td> |
| <td align="center">39.5</td> |
| <td align="center">58.7</td> |
| <td align="center">59.0 🥇</td> |
| <td align="center">56.1</td> |
| </tr> |
|
|
| <tr> |
| <td align="left">SWE-bench Pro</td> |
| <td align="center">61.2 🟢</td> |
| <td align="center">43.6</td> |
| <td align="center">59.6</td> |
| <td align="center">63.3</td> |
| <td align="center">64.6</td> |
| <td align="center">64.6 🥇</td> |
| </tr> |
|
|
| <tr> |
| <td align="left">SWE-bench Multilingual</td> |
| <td align="center">72.7 🟢</td> |
| <td align="center">58.7</td> |
| <td align="center">71.4</td> |
| <td align="center">80.8</td> |
| <td align="center">81.3 🥇</td> |
| <td align="center">74.1</td> |
| </tr> |
|
|
| <tr> |
| <td colspan="8" align="left"><b>📋 Instruction Following</b></td> |
| </tr> |
| |
| <tr> |
| <td align="left">IFBench</td> |
| <td align="center">81.6 🥇</td> |
| <td align="center">80.6</td> |
| <td align="center">44.6</td> |
| <td align="center">72.1</td> |
| <td align="center">53.4</td> |
| <td align="center">72.7</td> |
| </tr> |
|
|
| <tr> |
| <td align="left">AA-LCR</td> |
| <td align="center">81.0 🟢</td> |
| <td align="center">74.2</td> |
| <td align="center">73.2</td> |
| <td align="center">88.7 🥇</td> |
| <td align="center">79.7</td> |
| <td align="center">84.0</td> |
| </tr> |
|
|
|
|
| <tr> |
| <td colspan="8" align="left"><b>🧪 Scientific Research</b></td> |
| </tr> |
| |
| <tr> |
| <td align="left">FrontierScience-Olympiad</td> |
| <td align="center">81.0 🥇</td> |
| <td align="center">79.0</td> |
| <td align="center">70.0</td> |
| <td align="center">69.0</td> |
| <td align="center">67.0</td> |
| <td align="center">73.0</td> |
| </tr> |
|
|
| <tr> |
| <td align="left">FrontierScience-Research</td> |
| <td align="center">48.3 🥇</td> |
| <td align="center">40.0</td> |
| <td align="center">6.7</td> |
| <td align="center">21.7</td> |
| <td align="center">16.7</td> |
| <td align="center">26.7</td> |
| </tr> |
|
|
| <tr> |
| <td align="left">HiPhO</td> |
| <td align="center">45.3</td> |
| <td align="center">46.4 🥇</td> |
| <td align="center">39.5</td> |
| <td align="center">42.4</td> |
| <td align="center">39.7</td> |
| <td align="center">46.2</td> |
| </tr> |
|
|
| <tr> |
| <td align="left">HLE w/ tools</td> |
| <td align="center">55.1 🟢</td> |
| <td align="center">47.6</td> |
| <td align="center">33.4</td> |
| <td align="center">56.0</td> |
| <td align="center">62.5</td> |
| <td align="center">64.5 🥇</td> |
| </tr> |
|
|
| <tr> |
| <td align="left">CMT</td> |
| <td align="center">80.0 🥇</td> |
| <td align="center">72.0</td> |
| <td align="center">26.0</td> |
| <td align="center">34.0</td> |
| <td align="center">40.0</td> |
| <td align="center">50.0</td> |
| </tr> |
|
|
| <tr> |
| <td align="left">CritPt</td> |
| <td align="center">20.0 🟢</td> |
| <td align="center">4.3</td> |
| <td align="center">2.9</td> |
| <td align="center">23.4</td> |
| <td align="center">19.1</td> |
| <td align="center">32.3 🥇</td> |
| </tr> |
|
|
|
|
| <tr> |
| <td colspan="8" align="left"><b>🤖 General Agentic Tasks</b></td> |
| </tr> |
| |
| <tr> |
| <td align="left">Agents' Last Exam (ALE-CLI)</td> |
| <td align="center">19.1 🟢</td> |
| <td align="center">10.2</td> |
| <td align="center">14.3</td> |
| <td align="center">27.6</td> |
| <td align="center">28.5</td> |
| <td align="center">28.6 🥇</td> |
| </tr> |
|
|
| <tr> |
| <td align="left">VitaBench</td> |
| <td align="center">45.4 🟢</td> |
| <td align="center">38.7</td> |
| <td align="center">44.6</td> |
| <td align="center">59.0 🥇</td> |
| <td align="center">58.7</td> |
| <td align="center">46.8</td> |
| </tr> |
|
|
| <tr> |
| <td align="left">PinchBench</td> |
| <td align="center">90.0 🥇</td> |
| <td align="center">76.9</td> |
| <td align="center">88.6</td> |
| <td align="center">88.2</td> |
| <td align="center">86.6</td> |
| <td align="center">87.3</td> |
| </tr> |
|
|
| <tr> |
| <td colspan="8" align="left"><b>🔬 Scientific Agentic Tasks</b></td> |
| </tr> |
| |
| <tr> |
| <td align="left">ChemCOTBench</td> |
| <td align="center">66.5 🟢</td> |
| <td align="center">64.0</td> |
| <td align="center">53.8</td> |
| <td align="center">82.5 🥇</td> |
| <td align="center">78.3</td> |
| <td align="center">77.9</td> |
| </tr> |
|
|
|
|
| <tr> |
| <td align="left">MatTools</td> |
| <td align="center">42.6</td> |
| <td align="center">47.1 🟢</td> |
| <td align="center">38.6</td> |
| <td align="center">24.3</td> |
| <td align="center">53.2 🥇</td> |
| <td align="center">45.6</td> |
| </tr> |
|
|
| <tr> |
| <td align="left">OmniDocBench</td> |
| <td align="center">91.5 🥇</td> |
| <td align="center">89.5</td> |
| <td align="center">89.6</td> |
| <td align="center">91.1</td> |
| <td align="center">-</td> |
| <td align="center">85.8</td> |
| </tr> |
|
|
| </tbody> |
| </table> |
|
|
| > [!Note] |
| > SWE-Bench Series: We use the mini-SWE-agent harness with a temperature of 1.0, top-p of 0.95, a 200K-token context window, and automatic context compaction enabled. All evaluations are conducted in a fully offline environment without internet access. |
| > |
| > GLM-5.3 on OmniDocBench: GLM-5.3 is a text-only (unimodal) model without multimodal input support, so its results on OmniDocBench, a multimodal benchmark, are not reported. |
|
|
|
|
| ## Usage |
|
|
| ### SGLang |
|
|
| [SGLang](https://github.com/sgl-project/sglang) is a fast serving framework for large language models and vision language models. |
|
|
| Install SGLang with uv: |
|
|
| ```shell |
| uv venv --python 3.12 --seed --managed-python |
| source .venv/bin/activate |
| |
| uv pip install sglang |
| ``` |
|
|
| See [its documentation](https://docs.sglang.ai/get_started/install.html) for more details. |
|
|
| The following commands create API endpoints at `http://localhost:8000/v1`: |
|
|
| - **Standard Version**: |
|
|
| ```shell |
| python -m sglang.launch_server \ |
| --model-path InternScience/Sage-A1.5 \ |
| --port 8000 \ |
| --mem-fraction-static 0.8 \ |
| --context-length 262144 \ |
| --reasoning-parser qwen3 \ |
| --tool-call-parser qwen3_coder |
| ``` |
|
|
| - **Multi-Token Prediction (MTP)**: |
|
|
| ```shell |
| python -m sglang.launch_server \ |
| --model-path InternScience/Sage-A1.5 \ |
| --port 8000 \ |
| --mem-fraction-static 0.8 \ |
| --context-length 262144 \ |
| --reasoning-parser qwen3 \ |
| --tool-call-parser qwen3_coder \ |
| --speculative-algorithm EAGLE \ |
| --speculative-num-steps 3 \ |
| --speculative-eagle-topk 1 \ |
| --speculative-num-draft-tokens 4 |
| ``` |
|
|
| #### For ultra long context: |
|
|
| Sage-A1.5 natively supports a 256K context window. To extend it to 512K, serve the model with |
| YaRN (`factor: 2.0`) enabled by overriding the RoPE parameters: |
|
|
| ```shell |
| SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 \ |
| python -m sglang.launch_server \ |
| --model-path InternScience/Sage-A1.5 \ |
| --port 8000 \ |
| --context-length 524288 \ |
| --json-model-override-args '{ |
| "text_config": { |
| "rope_parameters": { |
| "mrope_interleaved": true, |
| "mrope_section": [11, 11, 10], |
| "rope_type": "yarn", |
| "rope_theta": 10000000, |
| "partial_rotary_factor": 0.25, |
| "factor": 2.0, |
| "original_max_position_embeddings": 262144 |
| } |
| } |
| }' \ |
| --reasoning-parser qwen3 \ |
| --tool-call-parser qwen3_coder \ |
| --mem-fraction-static 0.88 |
| ``` |
|
|
| ### vLLM |
|
|
| [vLLM](https://github.com/vllm-project/vllm) is a high-throughput and memory-efficient inference and serving engine for LLMs. |
|
|
| Install vLLM from the main branch via uv: |
|
|
| ```shell |
| uv venv --python 3.12 --seed --managed-python |
| source .venv/bin/activate |
| |
| uv pip install vllm --torch-backend=auto |
| ``` |
|
|
| See [its documentation](https://docs.vllm.ai/en/stable/getting_started/installation/index.html) for more details. |
|
|
| The following commands create API endpoints at `http://localhost:8000/v1`: |
|
|
| - **Standard Version** (1 GPUs, 262K context): |
|
|
| ```shell |
| vllm serve InternScience/Sage-A1.5 \ |
| --port 8000 \ |
| --max-model-len 262144 \ |
| --reasoning-parser qwen3 \ |
| --enable-auto-tool-choice \ |
| --tool-call-parser qwen3_coder |
| |
| ``` |
| - **Text-Only** (skips vision encoder to free KV cache memory): |
|
|
| ```shell |
| vllm serve InternScience/Sage-A1.5 \ |
| --port 8000 \ |
| --tensor-parallel-size 1 \ |
| --max-model-len 262144 \ |
| --reasoning-parser qwen3 \ |
| --language-model-only |
| ``` |
|
|
| - **Multi-Token Prediction (MTP)**: |
| ```shell |
| vllm serve InternScience/Sage-A1.5 \ |
| --port 8000 \ |
| --max-model-len 262144 \ |
| --reasoning-parser qwen3 \ |
| --enable-auto-tool-choice \ |
| --tool-call-parser qwen3_coder \ |
| --speculative-config '{"method":"mtp","num_speculative_tokens":3,"moe_backend":"triton"}' |
| ``` |
|
|
| #### For ultra long context: |
|
|
| Sage-A1.5 natively supports a 256K context window. To extend it to 512K, serve the model with |
| YaRN (`factor: 2.0`) enabled via `--hf-overrides` (vLLM no longer supports the old `--rope-scaling` |
| flag): |
|
|
| ```shell |
| VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 \ |
| vllm serve InternScience/Sage-A1.5 \ |
| --port 8000 \ |
| --max-model-len 524288 \ |
| --hf-overrides '{ |
| "text_config": { |
| "rope_parameters": { |
| "rope_type": "yarn", |
| "rope_theta": 10000000, |
| "partial_rotary_factor": 0.25, |
| "factor": 2.0, |
| "original_max_position_embeddings": 262144, |
| "mrope_interleaved": true, |
| "mrope_section": [11, 11, 10] |
| } |
| } |
| }' \ |
| --reasoning-parser qwen3 \ |
| --enable-auto-tool-choice \ |
| --tool-call-parser qwen3_coder |
| ``` |
|
|
| ### Advanced Modes |
|
|
| Sage-A1.5 ships two opt-in modes that build on the standard serving setup above. Both are off by |
| default: Agentic Orchestration Mode is turned on by registering a tool, Verification Mode by a chat-template flag. |
|
|
| #### Agentic Orchestration |
|
|
| Agentic Orchestration Mode (see [Highlights](#highlights)) lets the model fan out independent sub-questions to |
| sub-agents that run **in parallel**, each with its own context window and its own tool access. It is |
| enabled by exposing a `spawn_subagent` tool to the model: the model calls it whenever a task |
| decomposes into independent investigations, and your harness runs each sub-task and returns the |
| structured results. |
|
|
| Sub-agents are not restricted to any fixed tool set — they can be given whatever tools your |
| deployment registers (search, page reading, code execution, retrieval, internal APIs, …). By default |
| they inherit the main agent's tools; set `subagent_tools` if you want them to have a different, for |
| instance narrower, set. |
|
|
| Register the following tool schema alongside your other tools: |
|
|
| ```python |
| SPAWN_SUBAGENT_SCHEMA = { |
| "type": "function", |
| "function": { |
| "name": "spawn_subagent", |
| "description": ( |
| "Dispatch one or more sub-agents IN PARALLEL to investigate independent " |
| "sub-questions. Each sub-agent has access to google_search and read_page " |
| "but CANNOT see your conversation. You MUST give each sub-agent everything " |
| "it needs via the `task`, `context`, and `expected_output` fields. Sub-agents " |
| "return structured JSON: {answer, status, evidence, unresolved_issues}." |
| ), |
| "parameters": { |
| "type": "object", |
| "properties": { |
| "tasks": { |
| "type": "array", |
| "minItems": 1, |
| "description": "List of independent sub-tasks to dispatch in parallel.", |
| "items": { |
| "type": "object", |
| "properties": { |
| "task": { |
| "type": "string", |
| "description": "Self-contained sub-question. The sub-agent has no other context.", |
| }, |
| "context": { |
| "type": "string", |
| "description": "Background facts and constraints the sub-agent needs.", |
| }, |
| "expected_output": { |
| "type": "string", |
| "description": "Form of the desired answer (e.g. 'a year YYYY', 'a person name').", |
| }, |
| }, |
| "required": ["task", "context", "expected_output"], |
| }, |
| }, |
| }, |
| "required": ["tasks"], |
| }, |
| }, |
| } |
| ``` |
|
|
| Because a sub-agent cannot see the main conversation, every task must be self-contained — `task`, |
| `context`, and `expected_output` are all required. Each sub-agent returns one compact JSON object |
| `{answer, status, evidence, unresolved_issues}`, and the batch is handed back to the main model in |
| input order. |
|
|
| **Notes:** |
|
|
| - Keep each sub-task **independent**. Tasks in one batch run concurrently and cannot see each |
| other's results, so anything sequential belongs in separate `spawn_subagent` calls. |
| - `expected_output` acts as a format constraint on the sub-agent's `answer` (e.g. `'a year YYYY'`, |
| `'a person name'`), which keeps results short and easy to merge. |
| - A sub-agent that cannot resolve its task returns `"UNKNOWN"` with a `status` of `unknown` or |
| `insufficient_evidence` rather than guessing — check `status` before using an answer. |
| - Batch size and per-sub-agent turn budget are harness-side settings |
| (`max_tasks_per_spawn`, default 5; `subagent_max_turns`, default 20). Oversized batches are |
| rejected with an error message instead of being truncated. |
| - The schema's `description` above names `google_search` and `read_page` because that is the |
| reference setup. If you give sub-agents a different tool set, update that sentence to match — |
| it is what tells the main model what its sub-agents can actually do. |
|
|
| #### Verification Mode |
|
|
| Verification Mode (see [Highlights](#highlights)) is **opt-in** and off by default. Enable it per request by |
| passing `verification` through `chat_template_kwargs`; the chat template then appends the verification |
| protocol trigger to the end of your system message. |
|
|
| - **OpenAI Python client** (via `extra_body`): |
|
|
| ```python |
| from openai import OpenAI |
| |
| client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY") |
| |
| response = client.chat.completions.create( |
| model="InternScience/Sage-A1.5", |
| messages=[ |
| {"role": "system", "content": "You are a helpful assistant."}, |
| {"role": "user", "content": "Find the population of Beijing and cross-check it."}, |
| ], |
| tools=TOOLS, |
| extra_body={"chat_template_kwargs": {"verification": True}}, |
| ) |
| |
| message = response.choices[0].message |
| print(message.reasoning_content) # self-review happens inside the chain of thought |
| print(message.content) |
| ``` |
|
|
| - **curl**: |
|
|
| ```shell |
| curl http://localhost:8000/v1/chat/completions \ |
| -H 'Content-Type: application/json' \ |
| -d '{ |
| "model": "InternScience/Sage-A1.5", |
| "messages": [ |
| {"role": "system", "content": "You are a helpful assistant."}, |
| {"role": "user", "content": "Find the population of Shanghai and cross-check it."} |
| ], |
| "chat_template_kwargs": {"verification": true} |
| }' |
| ``` |
|
|
| **Notes:** |
|
|
| - A **non-empty system message is required**. Because the trigger is appended to the end of the system |
| content, a request with `verification: true` and no system message (or a whitespace-only one) is |
| rejected with HTTP 400. |
| - Pass a **JSON boolean**, not a string: `{"verification": true}`, not `{"verification": "true"}`. |
| - Omitting `chat_template_kwargs` is equivalent to `{"verification": false}`, which renders the prompt |
| unchanged. |
| - Verification Mode is independent of tool use — it works with or without `tools`. |
| - The self-review is emitted as reasoning content, so read `reasoning_content` in addition to `content`. |
|
|
| ### Recommended Sampling Parameters |
|
|
| For the best generation quality, we recommend the following sampling parameters: |
|
|
| - `temperature`: 1.0 |
| - `top_p`: 0.95 |
| - `top_k`: 20 |
| - `min_p`: 0.0 |
| - `presence_penalty`: 1.5 |
| - `repetition_penalty`: 1.0 |
| |
|
|
| ## Agent Capability Evaluation |
|
|
| To provide the community with a unified agent evaluation codebase for fair comparison, |
| we have also open-sourced an evaluation framework for assessing agentic models across core capabilities, including tool use and multi-step reasoning. |
| The evaluation code is included in the [Sage-A1/evaluation](https://github.com/InternScience/Agents-A1/tree/main/evaluation) of this repository. |
|
|
| We use this framework to evaluate the released model under a standardized and reproducible setting. |
| Specifically, the model is tested on a set of agent-oriented tasks that require it to understand user goals, decompose complex instructions, |
| interact with tools or environments when necessary, and produce final results. |
| The evaluation results reported in [Model Card](https://huggingface.co/InternScience/Sage-A1.5) are generated using the open-source framework above, |
| so that users can reproduce the experiments, compare other models under the same protocol, and further extend the benchmark for new agent scenarios. |
| (**Note that:** To ensure a fair comparison, we report the benchmark results from their original technical reports. |
| If a model does not report the corresponding benchmark results, we evaluate it using the same evaluation protocol as our model.) |
|
|
| For detailed evaluation scripts, task definitions, metrics, and reproduction instructions, please refer to the evaluation codebase. |
|
|
| ## Citation |
|
|
| If you find our work helpful, feel free to give us a cite. |
|
|
| ``` |
| @misc{lou2026sagea15longhorizonexecutionscientific, |
| title={Sage-A1.5: From Long-Horizon Execution to Scientific Exploration with Capability Composition}, |
| author={Sage-A1.5 Team}, |
| year={2026}, |
| eprint={2610.28578}, |
| archivePrefix={arXiv}, |
| primaryClass={cs.AI}, |
| url={https://arxiv.org/abs/2610.28578}, |
| } |
| ``` |
