Which one do I point my harness at?

Everything else, ranked

Capability score 0-100 · cuts from weights 2026.08.1

measured on Aider Polyglot, as of Jul 11, 2026

1. S80-1003

   Gemini 3 ProBest value

   Google

   88.7

   $0.28

   $9/M tokens·Tool use 89·Gemini CLI, Cursor, Aider

   GPT-5.1

   OpenAI

   87.9

   $0.39

   $15.20/M tokens·Tool use 85·Codex CLI, Cursor, Aider

   Claude Opus 4.6Best value

   Anthropic

   91.4

   $0.42

   $19/M tokens·Tool use 86·Claude Code, Cursor, Aider

   No model in this list works with that one yet.

2. A65-792

   DeepSeek V4Best value

   DeepSeek

   58.4

   $0.03

   $0.85/M tokens·Tool use 54·Aider, Cursor

   Manual call Jun 21, 2026

   Manual call B→A

   Benchmarks place it in B, but at 1/20th the price it is the only credible self-host path. Promoted with the reason on the record.

   Claude Sonnet 4.5Best value

   Anthropic

   78.2

   $0.21

   $11.40/M tokens·Tool use 74·Claude Code, Cursor, Aider

   No model in this list works with that one yet.

3. B45-642

   Mistral Large 3

   Mistral

   51.7

   $0.11

   Jun 4, 2026

   $4.80/M tokens·Tool use 52·Aider

   Grok 4.1

   xAI

   58.3

   No data

   $11.40/M tokens·Tool use No data·Cursor

   No model in this list works with that one yet.

4. C0-441

   Qwen3 72BBest value

   Alibaba

   38.9

   $0.02

   $0.32/M tokens·Tool use 42·Aider

   No model in this list works with that one yet.