qwen3.5-4b

evaluatedEdge/tiny

Vendor: AlibabaParams: 4BQuant: Q8Context: 128kLicense: Apache-2.0Type: local

Capability starfish

green = capability (pass-rate) · blue = operational (vs roster) · dashed = roster average

Overall

mean domain rate

Solved

60/77

latest per bench

Gauntlet time

14m

~21s/bench

Speed

43.4

tok/s

Steps/bench

1.7

decisiveness

$ / run

$0.0003

OpenRouter-equiv

Overall rank

#12 of 26

Agentic

#4 of 22

Coding

#19 of 22

Speed (tok/s)

#7 of 22

Decisive

#13 of 22

Cheapest

#6 of 9

Strongest

Gate · 100%

Weakest

Coding · 67%

Where it fits · live, recomputed from current results

#12 of 26 overall; strongest on Gate (100%); weakest on Coding (67%); #7/22 for speed (43.4 tok/s); #6/9 cheapest ($0.0003/run). (auto-generated from current results)

curator note: Punches far above 4B — 13/33 on the gauntlet once the parser bug was fixed. The strongest tiny model we have; viable on edge hardware and for fast, cheap first passes.

tinyfastedge

Verdict history · snapshotted daily when it changes

2026-07-09#12 of 26 overall; strongest on Gate (100%); weakest on Coding (67%); #7/22 for speed (43.4 tok/s); #6/9 cheapest ($0.0003/run). (auto-generated from current results)

2026-07-08#13 of 26 overall; strongest on Gate (100%); weakest on Coding (67%); #7/22 for speed (43.4 tok/s); #6/9 cheapest ($0.0003/run). (auto-generated from current results)

2026-07-07#14 of 28 overall; strongest on Gate (100%); weakest on SupportFlow (45%); #6/25 on the supportflow gauntlet; #8/26 for speed (39.4 tok/s); #5/9 cheapest ($0.0064/run). (auto-generated from current results)

2026-07-06#14 of 28 overall; strongest on Gate (100%); weakest on SupportFlow (45%); #8/24 on the supportflow gauntlet; #8/26 for speed (39.4 tok/s); #5/9 cheapest ($0.0064/run). (auto-generated from current results)

2026-07-05#14 of 28 overall; strongest on Gate (100%); weakest on SupportFlow (45%); #8/23 on the supportflow gauntlet; #8/26 for speed (39.4 tok/s); #5/9 cheapest ($0.0064/run). (auto-generated from current results)

2026-07-04#14 of 28 overall; strongest on Gate (100%); weakest on SupportFlow (47%); #6/23 on the supportflow gauntlet; #8/26 for speed (39.5 tok/s); #4/9 cheapest ($0.0047/run). (auto-generated from current results)

2026-07-03#14 of 28 overall; strongest on Gate (100%); weakest on SupportFlow (47%); #6/22 on the supportflow gauntlet; #8/26 for speed (39.5 tok/s); #4/9 cheapest ($0.0047/run). (auto-generated from current results)

2026-07-02#10 of 30 overall; strongest on Gate (100%); weakest on SupportFlow (42%); #7/26 on the supportflow gauntlet; #7/29 for speed (39.6 tok/s); #5/9 cheapest ($0.0047/run). (auto-generated from current results)

2026-07-01#6 of 28 overall; strongest on Gate (100%); weakest on SupportFlow (42%); #7/23 on the supportflow gauntlet; #8/27 for speed (39.6 tok/s); #3/9 cheapest ($0.0047/run). (auto-generated from current results)

2026-06-30#6 of 28 overall; strongest on Gate (100%); weakest on SupportFlow (40%); #7/21 on the supportflow gauntlet; #8/27 for speed (38.5 tok/s); #3/9 cheapest ($0.0035/run). (auto-generated from current results)

2026-06-29#5 of 28 overall; strongest on Gate (100%); weakest on SupportFlow (40%); #7/21 on the supportflow gauntlet; #8/27 for speed (38.5 tok/s); #3/9 cheapest ($0.0035/run). (auto-generated from current results)

Domain breakdown

Gate

100%

14/14

Coding

67%

16/24

Agentic

100%

14/14

73%

11/15

PHP

100%

5/5

Benches run

green = passed · red = failed · hover for harness & time

Gate (14)

Coding (24)

Agentic (14)

C# (15)

PHP (5)

SWE (5)

History (monthly pass-rate)

2026-06

87%

375/429

2026-07

50%

9/18

Availability record

When this model appeared in our catalogue census.

No census snapshots recorded yet.