IsAIokay.com
LiveUpdated 4 min ago
GitHub

The AI coding models developers trust right now.

Real developers set the ranking. No lab scores. No synthetic benchmarks.

Report from your terminal.

Install once, connect your tools, and submit only confirmed ratings on Linux, macOS, or Windows.

Open source and local-only. No code or prompts collected. View source ↗

isaiokay
  1. 01Install the CLI
    npm install --global @isaiokay/cli

    Same CLI via npm, pnpm, or Bun.

  2. 03Use your normal CLI
    codex

    Keep using Codex, Claude, Cursor, or another harness. Eligible questionnaires open after a session.

More setup options
Headless setup
isaiokay setup --headless

Open the link on any device. First-time auth still needs GitHub in a browser.

Connect everything detected
isaiokay install --all

Install every supported, unconfigured integration.

Preview without installing
npx --yes @isaiokay/cli --help

Also: pnpm dlx @isaiokay/cli --help or bunx @isaiokay/cli --help.

Coding models
RankModelDeveloper SignalTrend (7d)StatusConfidenceDevelopers
1
xAI logoGrok 4.6by xAIGrok model focused on long-running agents and ambitious interactive work
65/100
No move
Stable1 developer2 ratingsRate
2
Anthropic logoClaude Sonnet 5by AnthropicAnthropic model balancing speed and coding intelligence
62/100
No move
Stable1 developer2 ratingsRate
3
OpenAI logoGPT-5.6 Solby OpenAIFrontier GPT-5.6 model for complex professional and coding work
60/100
No move
Stable1 developer3 ratingsRate
4
DeepSeek logoDeepSeek V4 Flashby DeepSeekDeepSeek model optimized for agentic coding and tool use
62/100
No move
Stable1 developer1 ratingRate
5
Anthropic logoClaude Fable 5by AnthropicAnthropic model for long-running agents
Pending
Awaiting ratings
Collecting0 developers0 ratingsRate
6
Anthropic logoClaude Opus 5by AnthropicAnthropic model for complex agentic coding
Pending
Awaiting ratings
Collecting0 developers0 ratingsRate
7
Cursor logoComposer 2.5by CursorCursor coding model for complex instructions and long-running agent tasks
Pending
Awaiting ratings
Collecting0 developers0 ratingsRate
8
DeepSeek logoDeepSeek V4 Proby DeepSeekDeepSeek flagship model for coding, reasoning, and long-context agents
Pending
Awaiting ratings
Collecting0 developers0 ratingsRate
9
Google logoGemini 3.1 Proby GoogleGoogle preview model for complex agentic and coding work
Pending
Awaiting ratings
Collecting0 developers0 ratingsRate
10
Google logoGemini 3.5 Flashby GoogleGoogle model for agentic and coding workflows
Pending
Awaiting ratings
Collecting0 developers0 ratingsRate
11
Google logoGemini 3.6 Flashby GoogleStable Google model for fast agentic and multimodal work
Pending
Awaiting ratings
Collecting0 developers0 ratingsRate
12
Z.ai logoGLM-5.2by Z.aiZ.ai flagship model for long-horizon coding with a one-million-token contextClose
Pending
Awaiting ratings
Collecting0 developers0 ratingsRate
Recent experience evidence

Release window did not collect enough evidence. A possible regression is a community signal, never a claim about provider intent.

Rate GLM-5.2
Seven-day trendResult quality vs prior window: 0.0
Dimension scores
Result quality PendingUsage efficiency Pending
Usage pressure

Usage efficiency pending

How much useful progress developers felt they received for the subscription or usage consumed.

Release signal

Release window did not collect enough evidence

View release referenceOfficial model page
13
OpenAI logoGPT-5by OpenAIGeneral-purpose model used for coding tasks
Pending
Awaiting ratings
Collecting0 developers0 ratingsRate
14
OpenAI logoGPT-5.6 Lunaby OpenAIEfficient GPT-5.6 model for high-volume coding work
Pending
Awaiting ratings
Collecting0 developers0 ratingsRate
15
OpenAI logoGPT-5.6 Terraby OpenAIBalanced GPT-5.6 model for coding and everyday agentic work
Pending
Awaiting ratings
Collecting0 developers0 ratingsRate
16
Kimi logoKimi K2.7 Codeby KimiKimi model optimized for long-horizon coding and agent workflows
Pending
Awaiting ratings
Collecting0 developers0 ratingsRate
17
Kimi logoKimi K3by KimiKimi flagship coding model with native vision and long-context support
Pending
Awaiting ratings
Collecting0 developers0 ratingsRate
18
MiniMax logoMiniMax M3by MiniMaxMiniMax coding model with long context and native multimodal support
Pending
Awaiting ratings
Collecting0 developers0 ratingsRate
19
Qwen logoQwen 3.8 Maxby QwenQwen flagship model for long-horizon coding and professional work
Pending
Awaiting ratings
Collecting0 developers0 ratingsRate
12Z.ai logoGLM-5.2by Z.aiPending
Awaiting ratings
StatusCollectingDevelopers0Ratings0Confidence
Result quality Pending
Usage efficiency Pending

Release window did not collect enough evidence

Rate GLM-5.2
View full history for GLM-5.2
13OpenAI logoGPT-5by OpenAIPending
Awaiting ratings
StatusCollectingDevelopers0Ratings0Confidence
Result quality Pending
Usage efficiency Pending

No documented release baseline yet

Rate GPT-5
View full history for GPT-5
17Kimi logoKimi K3by KimiPending
Awaiting ratings
StatusCollectingDevelopers0Ratings0Confidence
Result quality Pending
Usage efficiency Pending

Release window did not collect enough evidence

Rate Kimi K3
View full history for Kimi K3

Scoring methodology

Developer Signal rewards a strong recent experience only when enough independent evidence supports it. The scoring specification lives here with the ranking it explains.

1

Weighted rating

Each eligible rating is normalized to a 0–100 scale, then combined across two deliberately distinct measurements.

Result quality
70%
Usage efficiency
30%
1 → 0 · 2 → 25 · 3 → 50 · 4 → 75 · 5 → 100
2

Confidence and breadth

Bayesian stabilization keeps sparse evidence cautious while independent developer participation strengthens the signal.

Evidence weight
recency × trust × (1 − fraud risk) × duplicate adjustment
Moderation
Excluded reports contribute nothing.
3

Recency and guardrails

Newer experience matters most, with safeguards for low-signal periods and documented model releases.

Live recency
2 ^ (−age in days / half-life days)
Half-life
Configurable 14 days, with no midnight reset.
Regression
Requires current confidence and a qualified release baseline.

72% weighted rating18% confidence10% breadth

Note: Only models are ranked. Coding agents such as Codex, Claude Code, Cursor, and OpenCode remain optional report context. Community ratings describe developer experience, not an objective benchmark or provider intent.

Want to rate Grok 4.6? Sign in with GitHub to report how the model felt and which coding agent you used.