Touchstone
← Models

gemma-4-26b-a4b-it

evaluatedCTO candidate
Vendor: GoogleParams: 26B-A4BActive: 4BQuant: Q8Context: 128kLicense: GemmaType: local
Capability starfish
green = capability (pass-rate) · blue = operational (vs roster) · dashed = roster average
Gate13/14Coding24/24Agentic12/14SupportFlow3/22C#15/15PHP5/5Speed22.3 tok/sDecisive1.5 steps/benchCost-eff$0.0001/run0
82%
Overall
mean domain rate
Solved
72/109
latest per bench
Gauntlet time
5m
~6s/bench
Speed
22.3
tok/s
Steps/bench
1.5
decisiveness
$ / run
$0.0001
OpenRouter-equiv
Overall rank
#15 of 26
Supportflow
#5 of 5
Agentic
#12 of 22
Coding
#2 of 22
Speed (tok/s)
#13 of 22
Decisive
#4 of 22
Cheapest
#1 of 9
Strongest
Coding · 100%
Weakest
SupportFlow · 14%
Where it fits · live, recomputed from current results

#15 of 26 overall; strongest on Coding (100%); weakest on SupportFlow (14%); #5/5 on the supportflow gauntlet; #13/22 for speed (22.3 tok/s); #1/9 cheapest ($0.0001/run). (auto-generated from current results)

curator note: Cheap and well-behaved on instructions. Mid-pack on hard multi-file fixes but reliable on gate/agentic. A strong value pick where cost-per-run matters more than peak capability.

moeinstructioncheap
Verdict history · snapshotted daily when it changes
2026-07-09#15 of 26 overall; strongest on Coding (100%); weakest on SupportFlow (14%); #5/5 on the supportflow gauntlet; #13/22 for speed (22.3 tok/s); #1/9 cheapest ($0.0001/run). (auto-generated from current results)
2026-07-08#16 of 26 overall; strongest on Coding (100%); weakest on SupportFlow (14%); #4/4 on the supportflow gauntlet; #13/22 for speed (22.3 tok/s); #1/9 cheapest ($0.0001/run). (auto-generated from current results)
2026-07-07#10 of 28 overall; strongest on Coding (100%); weakest on SupportFlow (39%); #9/25 on the supportflow gauntlet; #10/26 for speed (31.3 tok/s); #4/9 cheapest ($0.0064/run). (auto-generated from current results)
2026-07-06#10 of 28 overall; strongest on Coding (100%); weakest on SupportFlow (39%); #10/24 on the supportflow gauntlet; #10/26 for speed (31.3 tok/s); #4/9 cheapest ($0.0064/run). (auto-generated from current results)
2026-07-05#10 of 28 overall; strongest on Coding (100%); weakest on SupportFlow (39%); #11/23 on the supportflow gauntlet; #10/26 for speed (31.3 tok/s); #4/9 cheapest ($0.0064/run). (auto-generated from current results)
2026-07-04#9 of 28 overall; strongest on Coding (100%); weakest on SupportFlow (39%); #9/23 on the supportflow gauntlet; #10/26 for speed (31.3 tok/s); #5/9 cheapest ($0.0064/run). (auto-generated from current results)
2026-07-03#8 of 28 overall; strongest on Coding (100%); weakest on SupportFlow (44%); #8/22 on the supportflow gauntlet; #10/26 for speed (30.7 tok/s); #5/9 cheapest ($0.0065/run). (auto-generated from current results)
2026-07-02#5 of 30 overall; strongest on Coding (100%); weakest on SupportFlow (28%); #12/26 on the supportflow gauntlet; #12/29 for speed (30.7 tok/s); #6/9 cheapest ($0.0065/run). (auto-generated from current results)
2026-07-01#4 of 28 overall; strongest on Coding (100%); weakest on SupportFlow (32%); #10/23 on the supportflow gauntlet; #10/27 for speed (34.4 tok/s); #5/9 cheapest ($0.0063/run). (auto-generated from current results)
2026-06-29#4 of 28 overall; strongest on Coding (100%); weakest on SupportFlow (32%); #9/21 on the supportflow gauntlet; #10/27 for speed (34.4 tok/s); #5/9 cheapest ($0.0063/run). (auto-generated from current results)
Domain breakdown
Gate
93%
13/14
Coding
100%
24/24
Agentic
86%
12/14
SupportFlow
14%
3/22
C#
100%
15/15
PHP
100%
5/5
Benches run
green = passed · red = failed · hover for harness & time
Gate (14)
Coding (24)
Agentic (14)
SupportFlow (22)
C# (15)
PHP (5)
SWE (15)
History (monthly pass-rate)
2026-06
74%
351/476
2026-07
96%
111/116
Availability record
When this model appeared in our catalogue census.
2026-06-25 · openrouter2026-06-25 · openrouter