Hugging Face's logo

InternScience
/
Sage-A1.5

Text Generation
Transformers
Safetensors
qwen3_5_moe
image-text-to-text
Mixture of Experts
vlm
vision
agentic
code
tool-use
conversational
Model card Files
xet
 Community
9
Lonepic commited on
Commit
0e3205c
·
verified ·
1 Parent(s): ba7d931

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +24 -1
README.md CHANGED
@@ -97,7 +97,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
97
  <thead>
98
  <tr>
99
  <th rowspan="2" align="left">Benchmark</th>
100
- <th colspan="3" align="center" style="text-align:center;">
101
  📏 Comparable Models (~35B)
102
  </th>
103
 
@@ -109,6 +109,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
109
  <tr>
110
  <th align="center">Sage-A1.5</th>
111
  <th align="center">Agents-A1</th>
 
112
  <th align="center">Ornith-1.5</th>
113
 
114
  <th align="center">Kimi-K3</th>
@@ -128,6 +129,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
128
  <td align="left">GAIA</td>
129
  <td align="center">95.1</td>
130
  <td align="center">96.0 🥇</td>
 
131
  <td align="center">77.6</td>
132
  <td align="center">92.2</td>
133
  <td align="center">92.2</td>
@@ -138,6 +140,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
138
  <td align="left">BrowseComp</td>
139
  <td align="center">81.7 w/o Agentic Orchestration<br>82.8 🟢 w/ Agentic Orchestration</td>
140
  <td align="center">75.5</td>
 
141
  <td align="center">67.6</td>
142
  <td align="center">91.2 🥇</td>
143
  <td align="center">76.2</td>
@@ -148,6 +151,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
148
  <td align="left">DeepResearch-Bench2</td>
149
  <td align="center">43.7 w/o Agentic Orchestration<br>47.5 w/ Agentic Orchestration</td>
150
  <td align="center">37.5</td>
 
151
  <td align="center">48.6 🟢</td>
152
  <td align="center">52.4</td>
153
  <td align="center">55.7</td>
@@ -158,6 +162,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
158
  <td align="left">XBench-DS-2510</td>
159
  <td align="center">92.0 🥇</td>
160
  <td align="center">86.0</td>
 
161
  <td align="center">80.0</td>
162
  <td align="center">85.0</td>
163
  <td align="center">84.0</td>
@@ -174,6 +179,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
174
  <td align="left">MLE-Lite</td>
175
  <td align="center">65.2 🟢</td>
176
  <td align="center">43.9</td>
 
177
  <td align="center">39.4</td>
178
  <td align="center">72.7</td>
179
  <td align="center">72.7</td>
@@ -184,6 +190,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
184
  <td align="left">SciCode</td>
185
  <td align="center">44.3 🟢</td>
186
  <td align="center">44.3</td>
 
187
  <td align="center">39.5</td>
188
  <td align="center">58.7</td>
189
  <td align="center">59.0 🥇</td>
@@ -194,6 +201,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
194
  <td align="left">SWE-bench Pro</td>
195
  <td align="center">61.2 🟢</td>
196
  <td align="center">43.6</td>
 
197
  <td align="center">59.6</td>
198
  <td align="center">63.3</td>
199
  <td align="center">64.6</td>
@@ -204,6 +212,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
204
  <td align="left">SWE-bench Multilingual</td>
205
  <td align="center">72.7 🟢</td>
206
  <td align="center">58.7</td>
 
207
  <td align="center">71.4</td>
208
  <td align="center">80.8</td>
209
  <td align="center">81.3 🥇</td>
@@ -218,6 +227,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
218
  <td align="left">IFBench</td>
219
  <td align="center">81.6 🥇</td>
220
  <td align="center">80.6</td>
 
221
  <td align="center">44.6</td>
222
  <td align="center">72.1</td>
223
  <td align="center">53.4</td>
@@ -228,6 +238,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
228
  <td align="left">AA-LCR</td>
229
  <td align="center">81.0 🟢</td>
230
  <td align="center">74.2</td>
 
231
  <td align="center">73.2</td>
232
  <td align="center">88.7 🥇</td>
233
  <td align="center">79.7</td>
@@ -243,6 +254,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
243
  <td align="left">FrontierScience-Olympiad</td>
244
  <td align="center">81.0 🥇</td>
245
  <td align="center">79.0</td>
 
246
  <td align="center">70.0</td>
247
  <td align="center">69.0</td>
248
  <td align="center">67.0</td>
@@ -253,6 +265,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
253
  <td align="left">FrontierScience-Research</td>
254
  <td align="center">48.3 🥇</td>
255
  <td align="center">40.0</td>
 
256
  <td align="center">6.7</td>
257
  <td align="center">21.7</td>
258
  <td align="center">16.7</td>
@@ -263,6 +276,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
263
  <td align="left">HiPhO</td>
264
  <td align="center">45.3</td>
265
  <td align="center">46.4 🥇</td>
 
266
  <td align="center">39.5</td>
267
  <td align="center">42.4</td>
268
  <td align="center">39.7</td>
@@ -273,6 +287,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
273
  <td align="left">HLE w/ tools</td>
274
  <td align="center">55.1 🟢</td>
275
  <td align="center">47.6</td>
 
276
  <td align="center">33.4</td>
277
  <td align="center">56.0</td>
278
  <td align="center">62.5</td>
@@ -283,6 +298,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
283
  <td align="left">CMT</td>
284
  <td align="center">80.0 🥇</td>
285
  <td align="center">72.0</td>
 
286
  <td align="center">26.0</td>
287
  <td align="center">34.0</td>
288
  <td align="center">40.0</td>
@@ -293,6 +309,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
293
  <td align="left">CritPt</td>
294
  <td align="center">20.0 🟢</td>
295
  <td align="center">4.3</td>
 
296
  <td align="center">2.9</td>
297
  <td align="center">23.4</td>
298
  <td align="center">19.1</td>
@@ -308,6 +325,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
308
  <td align="left">Agents' Last Exam (ALE-CLI)</td>
309
  <td align="center">19.1 🟢</td>
310
  <td align="center">10.2</td>
 
311
  <td align="center">14.3</td>
312
  <td align="center">27.6</td>
313
  <td align="center">28.5</td>
@@ -318,6 +336,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
318
  <td align="left">VitaBench</td>
319
  <td align="center">45.4 🟢</td>
320
  <td align="center">38.7</td>
 
321
  <td align="center">44.6</td>
322
  <td align="center">59.0 🥇</td>
323
  <td align="center">58.7</td>
@@ -328,6 +347,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
328
  <td align="left">PinchBench</td>
329
  <td align="center">90.0 🥇</td>
330
  <td align="center">76.9</td>
 
331
  <td align="center">88.6</td>
332
  <td align="center">88.2</td>
333
  <td align="center">86.6</td>
@@ -342,6 +362,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
342
  <td align="left">ChemCOTBench</td>
343
  <td align="center">66.5 🟢</td>
344
  <td align="center">64.0</td>
 
345
  <td align="center">53.8</td>
346
  <td align="center">82.5 🥇</td>
347
  <td align="center">78.3</td>
@@ -353,6 +374,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
353
  <td align="left">MatTools</td>
354
  <td align="center">42.6</td>
355
  <td align="center">47.1 🟢</td>
 
356
  <td align="center">38.6</td>
357
  <td align="center">24.3</td>
358
  <td align="center">53.2 🥇</td>
@@ -363,6 +385,7 @@ efficient agentic model that narrows the gap with much larger frontier models.
363
  <td align="left">OmniDocBench</td>
364
  <td align="center">91.5 🥇</td>
365
  <td align="center">89.5</td>
 
366
  <td align="center">89.6</td>
367
  <td align="center">91.1</td>
368
  <td align="center">-</td>
 
97
  <thead>
98
  <tr>
99
  <th rowspan="2" align="left">Benchmark</th>
100
+ <th colspan="4" align="center" style="text-align:center;">
101
  📏 Comparable Models (~35B)
102
  </th>
103
 
 
109
  <tr>
110
  <th align="center">Sage-A1.5</th>
111
  <th align="center">Agents-A1</th>
112
+ <th align="center">Qwen3.6-35B-A3B</th>
113
  <th align="center">Ornith-1.5</th>
114
 
115
  <th align="center">Kimi-K3</th>
 
129
  <td align="left">GAIA</td>
130
  <td align="center">95.1</td>
131
  <td align="center">96.0 🥇</td>
132
+ <td align="center">78.6</td>
133
  <td align="center">77.6</td>
134
  <td align="center">92.2</td>
135
  <td align="center">92.2</td>
 
140
  <td align="left">BrowseComp</td>
141
  <td align="center">81.7 w/o Agentic Orchestration<br>82.8 🟢 w/ Agentic Orchestration</td>
142
  <td align="center">75.5</td>
143
+ <td align="center">67.9</td>
144
  <td align="center">67.6</td>
145
  <td align="center">91.2 🥇</td>
146
  <td align="center">76.2</td>
 
151
  <td align="left">DeepResearch-Bench2</td>
152
  <td align="center">43.7 w/o Agentic Orchestration<br>47.5 w/ Agentic Orchestration</td>
153
  <td align="center">37.5</td>
154
+ <td align="center">45.1</td>
155
  <td align="center">48.6 🟢</td>
156
  <td align="center">52.4</td>
157
  <td align="center">55.7</td>
 
162
  <td align="left">XBench-DS-2510</td>
163
  <td align="center">92.0 🥇</td>
164
  <td align="center">86.0</td>
165
+ <td align="center">71.0</td>
166
  <td align="center">80.0</td>
167
  <td align="center">85.0</td>
168
  <td align="center">84.0</td>
 
179
  <td align="left">MLE-Lite</td>
180
  <td align="center">65.2 🟢</td>
181
  <td align="center">43.9</td>
182
+ <td align="center">37.9</td>
183
  <td align="center">39.4</td>
184
  <td align="center">72.7</td>
185
  <td align="center">72.7</td>
 
190
  <td align="left">SciCode</td>
191
  <td align="center">44.3 🟢</td>
192
  <td align="center">44.3</td>
193
+ <td align="center">36.6</td>
194
  <td align="center">39.5</td>
195
  <td align="center">58.7</td>
196
  <td align="center">59.0 🥇</td>
 
201
  <td align="left">SWE-bench Pro</td>
202
  <td align="center">61.2 🟢</td>
203
  <td align="center">43.6</td>
204
+ <td align="center">49.5</td>
205
  <td align="center">59.6</td>
206
  <td align="center">63.3</td>
207
  <td align="center">64.6</td>
 
212
  <td align="left">SWE-bench Multilingual</td>
213
  <td align="center">72.7 🟢</td>
214
  <td align="center">58.7</td>
215
+ <td align="center">67.2</td>
216
  <td align="center">71.4</td>
217
  <td align="center">80.8</td>
218
  <td align="center">81.3 🥇</td>
 
227
  <td align="left">IFBench</td>
228
  <td align="center">81.6 🥇</td>
229
  <td align="center">80.6</td>
230
+ <td align="center">64.4</td>
231
  <td align="center">44.6</td>
232
  <td align="center">72.1</td>
233
  <td align="center">53.4</td>
 
238
  <td align="left">AA-LCR</td>
239
  <td align="center">81.0 🟢</td>
240
  <td align="center">74.2</td>
241
+ <td align="center">71.7</td>
242
  <td align="center">73.2</td>
243
  <td align="center">88.7 🥇</td>
244
  <td align="center">79.7</td>
 
254
  <td align="left">FrontierScience-Olympiad</td>
255
  <td align="center">81.0 🥇</td>
256
  <td align="center">79.0</td>
257
+ <td align="center">60.3</td>
258
  <td align="center">70.0</td>
259
  <td align="center">69.0</td>
260
  <td align="center">67.0</td>
 
265
  <td align="left">FrontierScience-Research</td>
266
  <td align="center">48.3 🥇</td>
267
  <td align="center">40.0</td>
268
+ <td align="center">2.9</td>
269
  <td align="center">6.7</td>
270
  <td align="center">21.7</td>
271
  <td align="center">16.7</td>
 
276
  <td align="left">HiPhO</td>
277
  <td align="center">45.3</td>
278
  <td align="center">46.4 🥇</td>
279
+ <td align="center">37.7</td>
280
  <td align="center">39.5</td>
281
  <td align="center">42.4</td>
282
  <td align="center">39.7</td>
 
287
  <td align="left">HLE w/ tools</td>
288
  <td align="center">55.1 🟢</td>
289
  <td align="center">47.6</td>
290
+ <td align="center">36.2</td>
291
  <td align="center">33.4</td>
292
  <td align="center">56.0</td>
293
  <td align="center">62.5</td>
 
298
  <td align="left">CMT</td>
299
  <td align="center">80.0 🥇</td>
300
  <td align="center">72.0</td>
301
+ <td align="center">20</td>
302
  <td align="center">26.0</td>
303
  <td align="center">34.0</td>
304
  <td align="center">40.0</td>
 
309
  <td align="left">CritPt</td>
310
  <td align="center">20.0 🟢</td>
311
  <td align="center">4.3</td>
312
+ <td align="center">0.3</td>
313
  <td align="center">2.9</td>
314
  <td align="center">23.4</td>
315
  <td align="center">19.1</td>
 
325
  <td align="left">Agents' Last Exam (ALE-CLI)</td>
326
  <td align="center">19.1 🟢</td>
327
  <td align="center">10.2</td>
328
+ <td align="center">3.81</td>
329
  <td align="center">14.3</td>
330
  <td align="center">27.6</td>
331
  <td align="center">28.5</td>
 
336
  <td align="left">VitaBench</td>
337
  <td align="center">45.4 🟢</td>
338
  <td align="center">38.7</td>
339
+ <td align="center">35.6</td>
340
  <td align="center">44.6</td>
341
  <td align="center">59.0 🥇</td>
342
  <td align="center">58.7</td>
 
347
  <td align="left">PinchBench</td>
348
  <td align="center">90.0 🥇</td>
349
  <td align="center">76.9</td>
350
+ <td align="center">88.1</td>
351
  <td align="center">88.6</td>
352
  <td align="center">88.2</td>
353
  <td align="center">86.6</td>
 
362
  <td align="left">ChemCOTBench</td>
363
  <td align="center">66.5 🟢</td>
364
  <td align="center">64.0</td>
365
+ <td align="center">59.3</td>
366
  <td align="center">53.8</td>
367
  <td align="center">82.5 🥇</td>
368
  <td align="center">78.3</td>
 
374
  <td align="left">MatTools</td>
375
  <td align="center">42.6</td>
376
  <td align="center">47.1 🟢</td>
377
+ <td align="center">15.9</td>
378
  <td align="center">38.6</td>
379
  <td align="center">24.3</td>
380
  <td align="center">53.2 🥇</td>
 
385
  <td align="left">OmniDocBench</td>
386
  <td align="center">91.5 🥇</td>
387
  <td align="center">89.5</td>
388
+ <td align="center">89.9</td>
389
  <td align="center">89.6</td>
390
  <td align="center">91.1</td>
391
  <td align="center">-</td>