Skip to content
Model Index
  • Overview
  • Models
  • Compare
  • Benchmarks
  • Finder
LabsMethodology

Model finder

Three questions, one ranked shortlist. Every recommendation shows the benchmark evidence and the cost behind it, and the scoring is spelled out at the bottom.

1. What is it for?

Coding agents, app building, refactors and migrations.

2. What matters more?

65% capability, 35% affordability.

3. Any hard requirements?

Start with GPT-6.1 Sol

18 models ranked for “build software”, balanced · 7 ruled out by your requirements · 15 too weak for this job · 1 without enough data.

  1. 1

    GPT-6.1 Sol

    Top match

    OpenAI

    Capability 86
    Affordability 40
    • 1757 on Arena WebDev · #4 of tracked models
    • 56.1% on Terminal-Bench 4.0 · #5 of tracked models
    • $0.72 measured cost per task · $2 in / $10 out per 1M
    • Cheaper at Medium effort: $0.21 per task for an index of 47.8 (vs 51.8)
  2. 2

    MiMo-V2.6-Pro

    Open weights

    Xiaomi

    Capability 69
    Affordability 69
    • 60.9% on SciCode · #5 of tracked models
    • 34.8% on Terminal-Bench 4.0 · #10 of tracked models
    • $0.13 measured cost per task · $0.43 in / $0.87 out per 1M
  3. 3

    Claude Opus 5.5

    Anthropic

    Capability 98
    Affordability 4
    • 1814 on Arena WebDev · #1 of tracked models
    • 66.9% on SciCode · #1 of tracked models
    • $5.98 measured cost per task · $4 in / $20 out per 1M
    • Cheaper at High effort: $1.82 per task for an index of 53.6 (vs 57.6)
  4. 4

    GPT-6 Astra

    OpenAI

    Capability 91
    Affordability 14
    • 1788 on Arena WebDev · #2 of tracked models
    • 59.1% on Terminal-Bench 4.0 · #3 of tracked models
    • $3.26 measured cost per task · $10 in / $50 out per 1M
    • Cheaper at Medium effort: $1.54 per task for an index of 49.6 (vs 52.7)
  5. 5

    Claude Sonnet 5.5

    Anthropic

    Capability 95
    Affordability 6
    • 63.6% on Terminal-Bench 4.0 · #1 of tracked models
    • 92.4% on Vibe Code · #1 of tracked models
    • $5.46 measured cost per task · $2 in / $10 out per 1M
    • Cheaper at xHigh effort: $2.01 per task for an index of 51.9 (vs 56.0)

How this ranking works

Capability for this use case is 85% coding and 15% reasoning & science, using the same 0–100 composite scores shown on every model page. Affordability is the measured cost per task on a log scale, where the cheapest tracked model scores 100; list price stands in when no measurement exists. 65% capability, 35% affordability. Models scoring under half the capability of the best eligible model are left out, so a near-free model never outranks a competent one.

It is a starting point, not a verdict. Read the methodology or sort the full list yourself.

Model Index

An independent snapshot of frontier SI models and how they perform on third-party benchmarks.

Data as of Oct 8, 2026

Explore

  • All models
  • Compare
  • Benchmarks
  • Model finder
  • Labs
  • Methodology & sources

Sources

Benchmark results come from Artificial Analysis, Arena, Vals AI and ARC Prize. Specifications and prices come from each lab's own documentation.

Not affiliated with any lab or evaluator. Names and logos are trademarks of their respective owners and are used only to identify them. Scores change often — check the linked source before making a decision that depends on one.

  • Overview
  • Models
  • Compare
  • Benchmarks
  • Finder
  • Overview
  • Models
  • Compare
  • Benchmarks
  • Finder
  • Overview
  • Models
  • Compare
  • Benchmarks
  • Finder