Jev Router
A model routing service based on the Jev decision model that selects a model and reasoning intensity for each request. It does not generate text and returns only typed answers with probabilities.
Timeline (last 90 days)
OpenRouter call records show that 117 of Jev Router's 118 calls were routed to GPT-6.1 Sol.
The author considers the comparison between Jev Router and GPT-6.1 Sol unclean, because the two are in different comparison groups, Sol's build runs were five days earlier and used an older goose build, and the router also selects reasoning effort for each call.
Jev Router's single call to GPT-6 Astra cost $1.80, accounting for 44.87% of the $4.02 total spend for build 1, and that call carried a 142,694-token prompt with zero cache hits.
Forge 2.0 requires each model to run two independent builds and ranks by median, and the leaderboard requires comparing models within the same comparison group; Jev Router and Gemini 4 Argon belong to comparison group 1, while Claude Opus 5.5 and GPT-6.1 Sol belong to comparison group 2.
In Forge 2.0, the median gap between two builds of a single model is 0.0832, Jev Router's two builds differ by 0.2125, and Gemini 4 Argon's two builds differ by 0.4983 (0.7926 vs. 0.2943).
TypeSafe's Jev Router scored 0.6398 and 0.8523 on two builds on the Forge 2.0 agentic coding benchmark, with a median of 0.746, ranking first.
Jev Router's low build of 0.6398 differs from GPT-6.1 Sol's high build of 0.6495 by only 0.0097, which under the rules of forge2_compare.py (a winner is decided only when all of one side's builds exceed all of the other's) should be judged "unresolved."
Goose sends reasoning: {effort: 'medium'} in all builds, but the OpenRouter documentation explicitly states that this is only a suggestion, not a guarantee, when going through Jev Router.
When GPT-6.1 Sol is called directly, reasoning tokens never exceed 516 across 95 calls; after routing through Jev Router, 19 of the 74 Sol calls in build 1 exceed 516, reaching as high as 3,327.
Jev Router's builds ran on goose 3.0.116, and GPT-6.1 Sol's builds ran on goose 3.0.110, 73 commits apart.
1 more timeline events
All Facts (11)
The author considers the comparison between Jev Router and GPT-6.1 Sol unclean, because the two are in different comparison groups, Sol's build runs were five days earlier and used an older goose build, and the router also selects reasoning effort for each call.
50%UnverifiedOpenRouter call records show that 117 of Jev Router's 118 calls were routed to GPT-6.1 Sol.
50%UnverifiedJev Router's single call to GPT-6 Astra cost $1.80, accounting for 44.87% of the $4.02 total spend for build 1, and that call carried a 142,694-token prompt with zero cache hits.
50%UnverifiedForge 2.0 requires each model to run two independent builds and ranks by median, and the leaderboard requires comparing models within the same comparison group; Jev Router and Gemini 4 Argon belong to comparison group 1, while Claude Opus 5.5 and GPT-6.1 Sol belong to comparison group 2.
50%UnverifiedIn Forge 2.0, the median gap between two builds of a single model is 0.0832, Jev Router's two builds differ by 0.2125, and Gemini 4 Argon's two builds differ by 0.4983 (0.7926 vs. 0.2943).
50%UnverifiedJev Router's low build of 0.6398 differs from GPT-6.1 Sol's high build of 0.6495 by only 0.0097, which under the rules of forge2_compare.py (a winner is decided only when all of one side's builds exceed all of the other's) should be judged "unresolved."
50%UnverifiedGoose sends reasoning: {effort: 'medium'} in all builds, but the OpenRouter documentation explicitly states that this is only a suggestion, not a guarantee, when going through Jev Router.
50%UnverifiedWhen GPT-6.1 Sol is called directly, reasoning tokens never exceed 516 across 95 calls; after routing through Jev Router, 19 of the 74 Sol calls in build 1 exceed 516, reaching as high as 3,327.
50%UnverifiedJev Router's builds ran on goose 3.0.116, and GPT-6.1 Sol's builds ran on goose 3.0.110, 73 commits apart.
50%UnverifiedOf the 118 calls Jev Router made in the benchmark, 117 went to GPT-6.1 Sol and only 1 went to GPT-6 Astra.
50%UnverifiedTypeSafe's Jev Router scored 0.6398 and 0.8523 on two builds on the Forge 2.0 agentic coding benchmark, with a median of 0.746, ranking first.
50%