AICODA
Compare real value

Find the AI coding subscription that delivers real value.

We evaluate 12 plans and 24 models across real-world coding benchmarks, usage limits, and reliability so you can choose with confidence.

▱
12
Plans
◇
24
Models
⌁
6
Benchmarks
◷
Updated
JUL 19, 2026
profile
Xiaomi MiMo

Xiaomi MiMo Token Plan

MiMo V2.5 Pro
66.4 value index
value
66.4 /100
coding
66.4 /100
reliability
64.2 /100
$6.00/mo · Lite

☆ Strong value performance makes Xiaomi MiMo Token Plan the current top pick.

Prices and credit totals are from the annual plan (effective USD/month and credits/month shown); a monthly-billing option (~12% higher) and a first-year "New User Offer" promo also exist. Credits are a shared pool across all 9 MiMo models. MiMo-V2.5-Pro ships as fully open MIT-licensed weights.
Z.ai

GLM Coding Plan

GLM-5.2
66.0 value index
value
66.0 /100
coding
66.0 /100
reliability
73.4 /100
$10/mo · Lite

☆ GLM Coding Plan stays close with balanced scores and credible coding depth.

Alibaba Qwen

Qwen Token Plan

Qwen3.6-27B
63.4 value index
value
63.4 /100
coding
63.4 /100
reliability
62.0 /100
$6.00/mo · Lite

☆ Competitive pricing and steady day-to-day performance keep Qwen Token Plan in the top three.

Coding Plan Pro is sold out for new subscribers; Alibaba recommends Token Plan, which includes Qwen3.8-Max-Preview (no independent coding benchmarks at launch). Limited-time Personal prices shown; confirm on the pricing page.

All plans 12 plans

PlanDetails
XI
Xiaomi MiMo Token Plan •
MiMo V2.5 Pro
$6.00
Lite
66.4 3/6
66.4
64.2
1M41.1 t/sOPEN
Z
GLM Coding Plan
GLM-5.2
$10
Lite
66.0 3/6
66.0
73.4
1M183.1 t/sOPEN
Q
Qwen Token Plan •
Qwen3.6-27B
$6.00
Lite
63.4 3/6
63.4
62.0
——OPEN
K
Kimi Membership + Kimi Code •
Kimi K3
$19
Moderato
73.0 3/6
57.1
58.1
1M62.0 t/s—
DS
DeepSeek API •
DeepSeek V4 Pro
$16
pay-per-token
66.3 3/6
55.5
70.4
1M53.2 t/sOPEN
CU
Cursor Pro •
Grok 4.5
$20
Pro
70.2 4/6
54.0
56.0
500K73.2 t/s—
M
MiniMax Token Plan •
MiniMax M3
$20
Plus
66.5 1/6
51.1
67.6
1.05M38.8 t/sOPEN
OP
opencode Go •
Kimi K3
$10
Go
over quota
73.0 3/6
50.8
59.5
1M62.0 t/s—
x
SuperGrok + Grok Build •
Grok 4.5
$30
SuperGrok
70.2 4/6
47.5
52.5
500K73.2 t/s—
◎
ChatGPT Pro + Codex •
GPT-5.6 Sol
$100
Pro 5x
74.4 4/6
37.2
72.9
1.05M——
AI
Claude Max •
Claude Fable 5
$100
Max 5x
72.2 3/6
36.1
75.3
1M60.7 t/s—
G
Google AI + Antigravity •
Gemini 3.1 Pro
$100
AI Ultra 5x
69.1 4/6
34.6
72.7
1M139.8 t/s—

ⓘ Scoring combines benchmark performance, real-world task success, usage limits, and price into Coding, Value, and Reliability (0–100).

Learn our methodology →

Methodology

Coding score

Difficulty-centered composite: each result counts as its deviation from that benchmark’s observed average, so being measured on a hard benchmark doesn’t hurt and a generously-scored one doesn’t inflate. Deviations are weighted over the full benchmark set — a missing benchmark counts at its average, so sparse coverage dilutes toward the mean — and anchored to the dataset average. The basis (e.g. “3/6”) shows coverage only, not the denominator. Self-reported results count at 75% of their deviation. Independent benchmarks weigh more than vendor-run ones. The Coding tab sorts by this score among plans whose estimated quota covers the selected usage profile; over-quota plans stay listed but rank below:

  • SWE-bench Verified ×1 mixed
  • GSO ×1 independent
  • Terminal-Bench 2.0 ×1.5 independent
  • DeepSWE ×1.5 independent
  • AA Coding Index ×1.5 independent
  • CursorBench ×0.5 vendor-self-reported

Value index

coding score ÷ log₁₀(effective cost), where effective cost is the cheapest tier whose estimated monthly quota covers your usage profile. Costs are floored at $10 so near-free API pricing can’t dominate the index. API-only cost assumes 80% input / 20% output tokens.

  • Light 5M tokens/mo
  • Daily 30M tokens/mo
  • Heavy 150M tokens/mo

Vendors express quotas in incompatible units (prompts per 5h, weekly hours, credits) — each tier carries its conversion assumption verbatim in the expanded row, and each plan a quota-confidence label (high/medium/low) for how solid the vendor’s published numbers are. The label never changes a score. Treat estimates as directional, not exact.

Reliability

70% curated editorial judgment per provider (incident/degradation history, quota pain, limit transparency — rationale in the expanded row) + 30% measured behavior of the shown model, the mean of three independent Artificial Analysis evals: non-hallucination rate (AA-Omniscience — admitting ignorance instead of confabulating), IFBench (instruction following) and τ²-Bench (agentic tool-calling dependability).

Every price, quota and benchmark entry links its source and carries a verification date. Data last verified Jul 19, 2026. Prices in USD as quoted by vendors; quarterly/annual plans converted to effective monthly.

© 2026 AICODA Legal notice Privacy policy

All scores are our evaluation and may change.