Model Directory — April 2026

Choose the right model stack,not just a model

27 models across four surfaces. Filter by job-to-be-done, assemble a lead + specialist + Remedy stack, and understand real credit costs before you ship.

View pricing

Want one answer

GLM-5-Turbo + MiMo V2 Pro

Balanced default stack

Cost-sensitive

MiMo V2 Flash

or Qwen3.6 Plus

Quality over speed

GPT-5.4

or Claude Sonnet 4.6

Remedy defaults

MiMo V2 Pro

Free chat: Qwen3.6 Plus

What are you optimizing for?

Balanced quality and cost for most engineering teamsBest starting point for most teams. Strong lead + reliable Remedy.

Routing templates

Recommended stacks

Pre-configured lead + specialist + Remedy combinations. Start with a preset, not a blank slate.

Best Default

Where most teams should start

3–8 cr / review
LeadGLM-5-Turbo
SpecialistQwen3.5-27B
RemedyMiMo V2 Pro
EscalationClaude Sonnet 4.6

Startup teams, balanced engineering orgs

Cheap Volume

Maximum coverage per credit spent

0.5–3 cr / review
LeadDeepSeek V3.2 Speciale
SpecialistQwen3.6 Plus
RemedyMiMo V2 Flash

High-volume orgs, cost-sensitive pipelines

Balanced Engineering

Strong defaults with escalation path

4–12 cr / review
LeadGLM-5.1
SpecialistKimi K2.5
RemedyGLM-5-Turbo
EscalationClaude Sonnet 4.6

Mid-size engineering teams, mixed PR sizes

Frontier Escalation

Standard stack + Ultra when it matters

20–60 cr / review
LeadClaude Sonnet 4.6
SpecialistGPT-5.4
RemedyGPT-5.1 Codex Max
EscalationClaude Opus 4.6

Frontier-only teams, high-stakes codebases

Remedy-First

Built for automated fix execution

3–14 cr / review
LeadMiMo V2 Pro
SpecialistQwen3.6 Plus
RemedyDeepSeek V3.2 Speciale
EscalationGPT-5.1 Codex Max

Teams shipping Remedy as a core workflow

Full catalog

Model explorer

27 models

Z.ai

GLM-5-Turbo

Default

Z.AI

Credit floor

3cr

Speed

Fast

Context

203K
Standard + ProLarge diffFree ChatRemedy

Best for: agent orchestration, long execution chains, stable tool use

XiaomiMiMo

MiMo V2 Pro

Default

Xiaomi

Credit floor

3cr

Speed

Balanced

Context

1M
Standard + ProLarge diffFree ChatRemedy

Best for: Remedy execution, agent adaptability, 1M context

Claude

Claude Sonnet 4.6

Default

Anthropic

Credit floor

22cr

Speed

Balanced

Context

1M
Standard + ProHigh-ambiguityRemedy

Best for: coding, codebase navigation, agents

OpenAI

GPT-5.4

Default

OpenAI

Credit floor

20cr

Speed

Balanced

Context

1M
Standard + ProHigh-ambiguityRemedy

Best for: coding, tool use, long-context reasoning

Qwen

Qwen3.6 Plus

Default

Alibaba

Credit floor

2cr

Speed

Balanced

Context

1M
Standard + ProLarge diffRemedy

Best for: repo-scale problem solving, coding, deep sub-agent work

Google

Gemma-4-31B

Google

Credit floor

0.5cr

Speed

Fast

Context

262K
Standard + ProMedium diffRemedy

Best for: broad low-cost coverage, document understanding, coding checks

XiaomiMiMo

MiMo V2 Flash

Xiaomi

Credit floor

0.5cr

Speed

Instant

Context

262K
Standard + ProPatchRemedy

Best for: reasoning at low cost, coding, agent loops

Stepfun

StepFun-3.5 Flash

Default

StepFun

Credit floor

1cr

Speed

Instant

Context

262K
Standard + ProPatchFree ChatRemedy

Best for: quick review passes, chat-like responsiveness, fast PR checks

KwaiKAT

KAT Coder Pro V2

KwaiPilot

Credit floor

1cr

Speed

Fast

Context

256K
Standard + ProMedium diffFree ChatRemedy

Best for: enterprise software tasks, coding-focused review, specialist work

Minimax

MiniMax-M2.5

MiniMax

Credit floor

1cr

Speed

Fast

Context

197K
Standard + ProMedium diffRemedy

Best for: general productivity, coding, efficient coverage

Minimax

MiniMax-M2.7

MiniMax

Credit floor

2cr

Speed

Balanced

Context

197K
Standard + ProMedium diffFree ChatRemedy

Best for: autonomous productivity, debugging, Remedy execution

Google

Gemini 3.1 Flash Lite

Google

Credit floor

2cr

Speed

Instant

Context

1M
Standard + ProPatch

Best for: high-volume tasks, broad coverage at minimal cost, fast sub-agent

OpenAI

GPT-5.4 Nano

OpenAI

Credit floor

2cr

Speed

Instant

Context

400K
Standard + ProPatchFree Chat

Best for: extraction, ranking, lightweight sub-agent work

Qwen

Qwen3.5-27B

Alibaba

Credit floor

2cr

Speed

Fast

Context

262K
Standard + ProMedium diffRemedy

Best for: balanced quality, fast responses, coding checks

DeepSeek

DeepSeek V3.2 Speciale

DeepSeek

Credit floor

2cr

Speed

Balanced

Context

164K
Standard + ProLarge diffRemedy

Best for: harder review tasks, Remedy execution, value reasoning

Kimi

Kimi K2.5

MoonshotAI

Credit floor

3cr

Speed

Balanced

Context

262K
Standard + ProLarge diffRemedy

Best for: multimodal review, agentic tasks, visual coding

Z.ai

GLM-5.1

Z.AI

Credit floor

4cr

Speed

Balanced

Context

203K
Standard + ProLarge diffRemedy

Best for: long-horizon coding, extended autonomous work, complex review

Google

Gemini 3 Flash

Google

Credit floor

4cr

Speed

Fast

Context

1M
Standard + ProLarge diff

Best for: near-Pro quality at lower latency, broad coverage, fast thinking

Qwen

Qwen3.5-397B-A17B

Alibaba

Credit floor

5cr

Speed

Balanced

Context

262K
Standard + ProLarge diffRemedy

Best for: stronger reasoning, code review, GUI tasks

OpenAI

GPT-5.4 Mini

OpenAI

Credit floor

6cr

Speed

Fast

Context

400K
Standard + ProMedium diffFree Chat

Best for: balanced capability + latency, general-purpose review, cost-aware teams

Grok

Grok 4.2

New

xAI

Credit floor

8cr

Speed

Fast

Context

2M
Standard + ProLarge diffFree Chat

Best for: fast flagship tasks, large context, strong tool-calling

OpenAI

GPT-5.1 Codex Max

OpenAI

Credit floor

14cr

Speed

Deep

Context

400K
Standard + ProHigh-ambiguityRemedy

Best for: long-running agentic coding, high-context software tasks, complex Remedy

Google

Gemini 3.1 Pro

Google

Credit floor

16cr

Speed

Deep

Context

1M
Standard + ProHigh-ambiguityRemedy

Best for: complex engineering, multimodal workflows, frontier reasoning

OpenAI

GPT-5.3 Codex

OpenAI

Credit floor

18cr

Speed

Deep

Context

400K
Standard + ProHigh-ambiguityRemedy

Best for: terminal-heavy tasks, SWE-style work, advanced coding review

Claude

Claude Opus 4.6

Anthropic

Credit floor

37cr

Speed

Deep

Context

1M
Ultra onlyHigh-ambiguityRemedy

Best for: large codebases, hard refactors, multi-step debugging

OpenAI

GPT-5.2 Pro

OpenAI

Credit floor

180cr

Speed

Deep

Context

400K
Ultra onlyHigh-ambiguityRemedy

Best for: high-stakes reasoning, coding at max depth, frontier analysis

OpenAI

GPT-5.4 Pro

OpenAI

Credit floor

237cr

Speed

Deep

Context

1M
Ultra onlyHigh-ambiguityRemedy

Best for: maximum OpenAI capability, complex system design review, frontier coding

Credit floor vs. real cost

Model floor is not the full review cost

The credit floor shows the minimum model cost lane. Actual review and Remedy spend depends on lead + specialist + depth multiplier. Remedy adds execution overhead including possible loop 2 re-review.

See full pricing and credit examples
Small PR< 50 lines changed

DeepSeek + MiMo Flash

1–4 cr
Normal PR50–300 lines changed

GLM-5-Turbo lead + Qwen specialist

3–12 cr
Large PR300+ lines, multi-file

Claude Sonnet + GPT-5.4

10–30 cr
Remedy runReview + automated fix loop

MiMo V2 Pro or GPT-5.1 Codex Max

+4–18 cr on top
Free for all accounts

Critique Chat has its own model picker

Critique Chat is free for signed-in users and does not spend PR review credits. The 8 chat models are separate from the main review and Remedy catalog.

Stepfun
StepFun-3.5 Flash
OpenAI
GPT-5.4-Nano
OpenAI
GPT-5.4-Mini
Z.ai
GLM-5-Turbo
Grok
Grok-4.2
XI
MiMo-V2-Pro
MI
MiniMax-M2.7
KwaiKAT
KAT Coder Pro V2
Transparency

How routing and access work

Automatic vs. pinned routing

By default, Critique selects models based on PR size, plan, and depth requirements. Power users and enterprise teams can pin specific models for lead, specialist, and Remedy roles.

How fallback works

If a model is unavailable or rate-limited, Critique falls back to the next appropriate model in the same cost tier. Ultra models do not fall back to Standard without explicit configuration.

Plan gates

Standard and Pro plans access the full catalog except Ultra-only models. Ultra plan unlocks Claude Opus 4.6, GPT-5.2 Pro, and GPT-5.4 Pro.

Raw model IDs

Raw OpenRouter IDs are shown in each expanded model card. They are secondary on this page but primary in API and admin contexts. Critique normalizes aliases automatically.

When to expose raw IDs

Use raw IDs when configuring CI workflows, calling the API directly, or pinning a specific model version. The dashboard UI uses display names.

Credits vs. real workloads

Credits are a normalized billing unit. 1 credit maps to a defined compute allocation. A full review cost is lead + specialist + any Remedy execution overhead, not just the floor.

Start with the best default stack

GLM-5-Turbo as lead, Qwen3.5-27B as specialist, MiMo V2 Pro for Remedy. Change any layer when you have data to justify it.