$

llm-rate

Friday, 21 August 2026

In the last 24 hours we dispatched 1,556 tasks across 4 models. Here's what we picked, and why.

What we ran

An autonomous AI fleet, written in TypeScript, picks a model per task using a complexity router. No vibes, no PR team. This is the actual production output of that router:

ModelDispatchesShareWhy this one
01 claude-sonnet-4-6 1,066 68.5% implementation (standard)
02 claude-haiku-4-5 278 17.9% implementation (light)
03 gpt-5.4-mini 189 12.1% implementation (codex pool)
04 claude-opus-4-6 23 1.5% implementation (high complexity)

Window: 24h to 2026-05-16T00:00:00Z. Source: daemon routing logs. The router writes a decision per dispatch; we parsed 1556 of them.

If you don't have a router, here are the picks per common task

Filtered from arena.ai's leaderboard plus published API prices. Filter thresholds are listed under each tab; arguable. Treat this as a starting shortlist, not a verdict.

Complex multi-step problems. Where to send the gnarly stuff.

Best value

mimo-v2.5

Xiaomi · quality 1438.2 · $0.24/M blended

Best quality

claude-opus-5-max

Anthropic · quality 1564.5 · $19.00/M blended

Filter: Math leaderboard, quality ≥ 1400. Sorted by quality. Use the cheaper end of this list as a router when you're sure the task is hard. 134 models survived.

Model Quality Ctx In /1M Out /1M Score ↓
01 claude-opus-5-maxqualityAnthropic 1564.5 1.0M $5.00 $25.00 1564.5
02 claude-opus-5-highAnthropic 1537.8 1.0M $5.00 $25.00 1537.8
03 gemini-3.7-flash-highGoogle 1527.4 1.0M $0.75 $3.57 1527.4
04 claude-fable-5Anthropic 1522.0 1.0M $10.00 $50.00 1522.0
05 claude-opus-4-6-highAnthropic 1515.9 1.0M $5.00 $25.00 1515.9
06 gemini-3.6-flash-highGoogle 1513.8 1.0M $0.38 $1.88 1513.8
07 claude-opus-4-6Anthropic 1511.9 1.0M $5.00 $25.00 1511.9
08 gemini-3.5-flash-highGoogle 1508.2 1.0M $0.75 $4.50 1508.2
09 qwen3.8-maxAlibaba 1507.8 1.0M $2.00 $6.00 1507.8
10 claude-opus-4-7-highAnthropic 1498.0 1.0M $5.00 $25.00 1498.0
11 qwen3.7-max-previewAlibaba 1492.7 1.0M $1.48 $4.42 1492.7
12 kimi-k3-maxMoonshot 1492.2 1492.2

What this is, and isn't

Right now this is filter-on-arena.ai plus a public log of what we ran. Arena Elo measures pairwise human preference on short prompts. It does not measure: whether a model produces valid JSON under a schema, whether it hallucinates function names, whether it refuses queries it shouldn't, latency p99, rate-limit behaviour. Production teams need those signals.

We're building a benchmark runner — fixed prompt suites for RAG, structured extraction, code refactoring, function calling — run daily against every model. Raw inputs, outputs, judge rationale, costs published. When that lands, the "picks" section gets its real backing. Until then, the picks section is opinion with a citation, not measurement.