30% off open-source modelsends in 11d 10:45:31
DevPassby LLM Gateway
Census open · Q3 wave filing now

Which coding models are actually worth the money?

The 2026 DevPass Model Census: value, quality, and speed — rated only by developers who shipped with these models. Every rating is backed by at least 50 real requests through LLM Gateway in the past 30 days. No benchmarks, no vibes.

File your entry · claim a free Reset PassGet your DevPass
767
Entries filed
624
Developers reporting
23
Models on the registry

The registry · ranked by value score

Doc. CS-2026
1
qwen3.8-max
5 verified ratings · mostly Agentic coding
100%
recommend
Value
4.8
Quality
4.4
Speed
3.2
2
gemini-3.7-flash
5 verified ratings · mostly Writing tests
100%
recommend
Value
4.6
Quality
4.2
Speed
4.8
3
deepseek-v4-flash
43 verified ratings · mostly Agentic coding
100%
recommend
Value
4.6
Quality
3.8
Speed
4.3
4
deepseek-v4-pro
49 verified ratings · mostly Agentic coding
98%
recommend
Value
4.5
Quality
3.7
Speed
3.9
5
gemini-3-flash-preview
31 verified ratings · mostly Writing tests
100%
recommend
Value
4.3
Quality
3.8
Speed
3.8
6
gpt-5.1
12 verified ratings · mostly Writing tests
83%
recommend
Value
4.2
Quality
4.2
Speed
3.7
7
gpt-5.6-sol
171 verified ratings · mostly Writing tests
95%
recommend
Value
4.1
Quality
4.3
Speed
3.8
8
gemini-3.6-flash
85 verified ratings · mostly Writing tests
93%
recommend
Value
4.1
Quality
3.8
Speed
4.3
9
claude-sonnet-5
15 verified ratings · mostly Agentic coding
87%
recommend
Value
4.0
Quality
4.1
Speed
3.7
10
gpt-5.6-luna
20 verified ratings · mostly Agentic coding
85%
recommend
Value
4.0
Quality
3.9
Speed
4.0
11
gemini-3.1-pro-preview
111 verified ratings · mostly Other
87%
recommend
Value
3.9
Quality
3.7
Speed
3.4
12
glm-5.2
50 verified ratings · mostly Agentic coding
96%
recommend
Value
3.9
Quality
4.0
Speed
3.4
13
gemini-3.5-flash
17 verified ratings · mostly Writing tests
88%
recommend
Value
3.8
Quality
3.8
Speed
3.9
14
gpt-5.5
9 verified ratings · mostly Agentic coding
89%
recommend
Value
3.8
Quality
4.2
Speed
3.8
15
gemini-3.5-flash-lite
9 verified ratings · mostly Code review
78%
recommend
Value
3.8
Quality
3.6
Speed
4.1
16
claude-opus-4-6
47 verified ratings · mostly Writing tests
98%
recommend
Value
3.8
Quality
4.2
Speed
3.7
17
kimi-k3
27 verified ratings · mostly Agentic coding
96%
recommend
Value
3.6
Quality
4.3
Speed
4.0
18
gpt-5.6-terra
10 verified ratings · mostly Agentic coding
70%
recommend
Value
3.5
Quality
3.4
Speed
3.8
19
minimax-m3
6 verified ratings · mostly Agentic coding
83%
recommend
Value
3.5
Quality
3.5
Speed
3.2
20
claude-fable-5
14 verified ratings · mostly Agentic coding
93%
recommend
Value
3.4
Quality
4.4
Speed
3.3
21
grok-4-5
11 verified ratings · mostly Agentic coding
91%
recommend
Value
3.4
Quality
3.7
Speed
3.9
22
claude-opus-5
9 verified ratings · mostly Agentic coding
89%
recommend
Value
3.3
Quality
3.9
Speed
3.4
23
claude-opus-4-8
11 verified ratings · mostly Agentic coding
91%
recommend
Value
3.0
Quality
4.2
Speed
3.8

Scores are 1–5 averages. Models need 5+ verified ratings to be listed; totals refresh every few minutes.

The rules of the registry

01
Usage-verified, or it doesn't count

You can only rate a model your DevPass workspace has hit with 50+ requests in the past 30 days. Nobody rates a model they read a thread about.

02
Members only

Every respondent has an active, paid DevPass plan. These are verdicts from people spending their own credits.

03
No small-sample noise

A model is published only after 5 or more developers rate it, and only aggregates ever leave the building.

04
One reward per member per quarter

The census runs in quarterly waves. Your first entry of each wave earns a free Reset Pass — rate as many models as you use, but nobody can farm passes.

Rate your models · get a free Reset Pass