Touchstone

Models

Every model under test — local and cloud. Coverage is the share of the full gauntlet attempted; Ctx is the peak context the model actually consumed (flags models that burn context differently). Click any model for its full profile, capability starfish, and history.

ModelOverallCoverageSolvedSpeedCtxQuantLast runBest atStatus
claude-sonnet-4.6
· cloud
97%
43/65
46/4823.8 t/s6.4%cloud2026-07-06gateevaluated
deepseek-r1-32b
80%
5/65
4/102026-07-02php
deepseek-r1-distill-qwen-32b
DeepSeek
53%
43/65
27/485 t/s19.1%Q8_02026-07-03codingevaluated
devstral-small-2-2512
Mistral AI · 24B
81%
46/65
43/628.2 t/s18.6%Q82026-06-23gateevaluated
gemini-2.5-flash-lite
· cloud
74%
43/65
36/4826 t/s19.8%cloud2026-07-01gateevaluated
gemma-4-26b-a4b-it
Google · 26B-A4B
82%
65/65
72/10922.3 t/s19.9%Q82026-07-01codingevaluated
gpt-oss-120b
OpenAI
92%
43/65
45/4847.1 t/s17.6%MXFP42026-06-17gateevaluated
gpt-oss-20b
OpenAI · 20B
96%
43/65
73/10072.1 t/s17.7%MXFP42026-07-01codingevaluated
granite-4.1-30b
IBM · 30B
96%
43/65
47/507.1 t/s18.8%Q82026-07-02gateevaluated
granite-4.1-8b
IBM
77%
43/65
38/5920.5 t/s18.8%Q82026-06-23gateevaluated
hermes-4.3-36b
18%
22/65
4/222026-07-08supportflow
llama-3.3-70b
Meta
94%
41/65
39/532.8 t/s17.5%Q82026-06-23codingevaluated
mistral-small-3.1-24b-instruct-2503
Mistral AI · 24B
86%
43/65
43/598.2 t/s18.6%Q82026-06-23gateevaluated
phi-4
100%
5/65
5/102026-07-02php
phi-4-reasoning-plus
Microsoft
40%
43/65
19/4814.1 t/s50.6%Q82026-07-02codingevaluated
Qwen3-235B-A22B (cluster)
235B-A22B
90%
20/65
17/20Q42026-06-30phpshelved
qwen3-coder-30b
Alibaba · 30B-A3B
75%
43/65
50/9330.8 t/s18.6%Q82026-07-01gateevaluated
qwen3-next-80b
Alibaba · 80B-A3B
82%
58/65
53/7934.3 t/s18.6%Q82026-07-02codingactive
qwen3.5-0.8b
Alibaba
29%
43/65
20/5484.7 t/s19.2%Q82026-07-02agenticevaluated
qwen3.5-122b-a10b-cluster
89%
43/65
43/4816.3 t/s21.6%Q8_02026-06-29gate
qwen3.5-2b
Alibaba
53%
43/65
31/5468.9 t/s19.2%Q82026-07-02agenticevaluated
qwen3.5-4b
Alibaba · 4B
88%
43/65
60/7743.4 t/s20.9%Q82026-07-02gateevaluated
qwen3.6-27b
Alibaba
96%
43/65
85/907.8 t/s19.8%Q8_02026-07-05gateevaluated
qwen3.6-35b-a3b
Alibaba · 35B-A3B
88%
65/65
80/12246 t/s22.9%Q82026-07-05codingin-prod
qwen3.6-35b-a3b
Alibaba · 35B-A3B
95%
43/65
46/4833.1 t/s20.9%BF162026-07-02gateevaluated
qwen3.6-35b-a3b
Alibaba · 35B-A3B · cloud
97%
43/65
46/4897.3 t/s21.4%cloud2026-07-01gateevaluated