MODEL VERSIONS ON THIS PAGE HAVE MOVED — CHECKED 8 OCTOBER 2026
Current line, with published API pricing per million tokens:
●
OpenAI — GPT-6 Sol (paid ChatGPT), GPT-6 Luna ($0.10 / $0.50), GPT-6.1 Sol ($2.00 / $10.00), GPT-6 Astra
●
Anthropic — Claude Opus 5.5, Sonnet 5.5 ($2.00 / $10.00), Haiku 5.5 ($0.10 / $0.50 below 100K), Fable 5.1
●
xAI — Grok 4.3 ($1.25 / $2.50), Grok 4.7 ($2.00 / $6.00)
●
Google — Gemini 3.5 Flash-Lite ($0.30 / $2.50), Gemini 3.8 Flash ($0.75 / $3.75 to 31 Dec), Gemini 4 Argon
The comparison below is kept because the reasoning still holds — how these tiers trade price against capability has not changed. The version numbers and prices have.
Current pricing table →
QUICK VERDICT — JULY 26, 2026 (K3 WEIGHTS TONIGHT)
● Better agentic coding: Kimi K3 — SWE Marathon 42.0% #1
● Better reasoning (ARC-AGI-3): Claude Opus 5 — 30.2% (K3 not published)
● Cheaper API: Kimi K3 — $3/$15/M vs Opus 5 $5/$25/M
● Open weights: Kimi K3 only — live tonight at 8PM ET
● Data residency (hosted API): Opus 5 — no concern. K3 — China NI Law (resolved via self-hosting after tonight)
● Hallucination risk: K3 — 51% rate from independent testing (not in Moonshot benchmarks). Opus 5 — higher than Opus 4.8 on one factual benchmark per Anthropic.
● Context: Both 1M tokens
Kimi K3 for: Agentic coding (SWE Marathon #1), frontend development (Design Arena #1), cost-sensitive workloads (40% cheaper API), and teams who want self-hosted inference with data sovereignty after tonight's weight release. Test hallucination on your use case first.
Claude Opus 5 for: Writing quality, factual accuracy tasks (better hallucination profile), ARC-AGI-3 reasoning, effort-level control, global availability with no data residency concern on the hosted API, and production workloads where Anthropic's auto-fallback matters.
Last updated July 26, 2026. Related: K3 weights tonight — checklist → · Opus 5 full review →