📍 Selection Hub · Updated 2026-08

How to Choose an AI Coding Model
in H2 2026: The Full Picture

Not "which is strongest" — "which fits your situation." A three-axis decision framework (budget, task complexity, context needs), a snapshot of all three camps, and a single hub for this site's lineup content.

#Decision Framework#Three Camps#Navigation#2026-08
Current state: all three camps have shipping flagships, one is still rumored

Claude (Opus 5 / Fable 5) and OpenAI (the three GPT-5.6 tiers) are all live on QCode; the domestic open-weight lines — GLM-5.3 (08-14), Kimi K3 (07-16, weights open-sourced 07-27) and DeepSeek V4 Pro (GA 08-13, peak/off-peak pricing from 08-17) — are all on sale. Google's Gemini 3.5 Pro remains in limited preview, with rumored dates already missed twice.

Landscape as of: 2026-08-22.

Decision Framework: Three Axes

Answer these three questions before comparing benchmark scores — it's more effective.

Axis Ask yourself Lean toward Typical scenario Note
Budget sensitivity Is this workload large enough that cost control matters? Tight budget → Luna / Sonnet 5; flexible → Sol / Fable 5 Bulk completion, CI scripts, light classification Luna is priced at roughly 1/25 of Sol — the gap matters at high volume
Task complexity Is this a long-horizon refactor or high-difficulty reasoning task, or routine feature work? High difficulty → Sol / Fable 5; routine → Terra / Sonnet 5 Cross-repo refactors, complex debugging, architecture decisions On the hardest slice of tasks, the flagship success-rate gap far outweighs the price gap
Context needs How much context does a single task need? Fable 5 (1M) / Sol・Terra・Luna (1.05M) / Gemini 3.5 Flash (1.05M) Long-document Q&A, large cross-file refactors Current flagship context windows across all three camps are similar — not the main differentiator

Three-Camp Snapshot

Claude (Anthropic)

Opus 5: the current flagship (2026-07-24, $5/$25, 1M context). Fable 5: launched 06-09, $10/$50, SWE-bench Pro 80.3 official methodology / 80.0 aggregated board. Sonnet 5: the balanced workhorse, available on QCode. Opus 4.8: previous-generation flagship, still on sale.

OpenAI

GPT-5.6 three tiers: Sol (flagship $5/$30), Terra (balanced $2/$12), Luna (fast $0.20/$1.20), cut on 07-30. GPT-5.3-Codex left in 2026-08, official migration to the 5.6 tiers. Official wording (rechecked 2026-08-30 still “starting August 31”): from 2026-08-31 GPT-5.4 leaves ChatGPT-signed-in Codex (API-key access is not in this retirement). There is also permissioned GPT-5.6-Cyber (Daybreak Red; not on ordinary channels).

Google

Gemini 3.6 Flash (07-21) and 3.7 Flash (GA 08-13, intro price $0.75/$3.75 through 2026-12-31) have shipped; Gemini 3.5 Pro is still unreleased — both previously rumored dates (July and 08-12) have slipped, Google has confirmed no date, so don't treat it as a done deal.

GPT-5.4 Retirement Migration Guide

Before the 8/31 Codex retirement: should you move to Sol, Terra, or Luna?

Claude Fable 5 vs GPT-5.6 Sol

Specs, pricing, and SWE-Bench Pro scores for both flagships, side by side.

What Is the Relationship Between Claude Mythos 5 and Fable 5?

What everyday users can actually access is Fable 5 — Mythos 5 has no public commercial access.

Recent Timeline

Key milestones across all three camps over the past two months.

2026-07-09

GPT-5.6's three tiers fully launch

Sol, Terra, and Luna open to everyone via API and Codex; QCode adds support the same day.

2026-07-30

GPT-5.6 price cuts

Luna cut 80% to $0.20/$1.20, Terra cut 20% to $2/$12, Sol unchanged.

2026-08-14

August's packed release week

GLM-5.3 released (same base as 5.2, post-training scaling, coding +50%); Qwen3.8-Max launched 08-03 and open-sourced Max-tier weights for the first time on 08-12; GPT-5.6-Cyber entered the vetted Daybreak Red tier on 08-10.

2026-08-31

GPT-5.4 retires from Codex

Official wording: ChatGPT-signed-in sessions only. API / Codex-key users are outside this retirement. Rechecked 2026-08-30, official text was still “starting August 31”.

About this page

This page is a decision-framework and content-navigation hub — it doesn't re-verify each model's primary facts. See the linked deep-dive pages for exact figures and sources. Pricing is sourced from the LiteLLM open pricing JSON.

FAQ

What's the first thing to check when picking an AI coding model today?

Don't jump straight to benchmark scores. Answer three questions first: budget sensitivity, task complexity, and per-task context needs. Once those are answered, the candidate list is usually down to 1-2 tiers, and benchmark scores can settle the final call.

Do all GPT-5.4 users need to migrate?

No. The official line is that the 2026-08-31 cutoff covers ChatGPT-signed-in sessions only (rechecked 2026-08-30, still “starting August 31”). API or Codex-key access is outside this announcement; you can keep using it or move to the matching GPT-5.6 tier.

Is Gemini 3.5 Pro available yet?

Not yet. As of 2026-08-22 it's still in limited preview; both previously rumored launch dates (July and August 12) have slipped, and Google has confirmed no new date — treat it as a rumor, not a basis for decisions.

How is this page different from the 'AI Model Radar'?

The radar page is a snapshot of which models are available/preview/restricted. This page is a decision framework for which model fits your situation, plus a hub for this site's new and updated deep-dive content. They're complementary — read both.

Create a QCode account

Plans from ¥60 / $8.57. One key across Claude, GPT, and Gemini models.

Related Reading

Prices, benchmark scores, and launch dates on this page have been cross-checked; the Gemini 3.5 Pro launch date is an unconfirmed rumor. Information may change — always defer to each vendor's official announcements.