Skip to main content

​
Key Takeaways

  • Three tiers at once: OpenAI released the GPT-5.6 series on July 9, 2026 — flagship gpt-5.6-sol, balanced gpt-5.6-terra, and lightweight gpt-5.6-luna — and all three are live on APIYI
  • New naming system: the number marks the model generation while Sol / Terra / Luna are durable capability tiers that advance on their own cadence, replacing the old Pro / Mini / Nano suffixes
  • Clear gains over GPT-5.5: Sol hits 88.8% on Terminal-Bench 2.1 (GPT-5.5: 85.6%), 90.4% on BrowseComp, and 52.7% on Agents’ Last Exam; Terra matches GPT-5.5-level performance at half the price
  • Programmatic Tool Calling: the Responses API gains programmatic tool calling — model-written JavaScript runs in an isolated, network-less V8 runtime, with customers reporting 38-63.5% token reductions
  • Dual-channel supply: the default official-relay group is at official price parity (recharge bonus up to 20%, roughly 17% off effective); the CodexReverse group carries a default 0.7x discount

​
Background

On July 9, 2026, OpenAI moved the GPT-5.6 series from limited preview to general availability across ChatGPT, Codex, and the OpenAI API. The headline change is the naming system: instead of Pro / Mini / Nano suffixes, the family introduces three celestial capability tiers — Sol (flagship), Terra (balanced), Luna (lightweight) — where the number identifies the generation and each tier can evolve on its own schedule. The positioning is equally crisp: Sol pushes the frontier on agentic tasks, Terra delivers GPT-5.5-class performance at half the cost, and Luna drops the entry price to $1/$6 per million tokens. APIYI shipped all three models the day after release, on both the default official-relay group and the CodexReverse group.

​
Deep Dive

​
Benchmarks (data retrieved 2026/7/10)

BenchmarkSolTerraLunaGPT-5.5Claude Fable 5
AA Coding Agent Index v1.18077.474.676.477.2
Terminal-Bench 2.188.8%87.4%84.7%85.6%83.1%
DeepSWE v1.172.7%69.6%67.2%67%69.7%
SWE-Bench Pro64.6%63.4%62.7%59.4%80%
Agents’ Last Exam52.7%50.4%50.3%46.9%40.5%
BrowseComp90.4%87.5%83.3%84.4%—
OSWorld 2.062.6%50.2%45.6%47.5%—
Sources: the official OpenAI announcement at openai.com/index/gpt-5-6 and marktechpost.com (July 9, 2026). The multi-agent Sol Ultra configuration (4 agents) reaches 91.9% on Terminal-Bench 2.1. Note: Claude Fable 5 (80%) still leads Sol (64.6%) by a wide margin on SWE-Bench Pro.

​
Core Features

Three-Tier Capability System

Sol for the ceiling, Terra for the daily workhorse, Luna for cost control — same generation and lineage, switchable by just changing the model name in the OpenAI-compatible format

Programmatic Tool Calling

New in the Responses API: model-written JavaScript executes in an isolated, network-less V8 runtime, with named customers reporting 38-63.5% token reductions

Upgraded Cache Billing

Cache writes bill at 1.25x the uncached input rate; cache reads keep the 90% discount with a 30-minute minimum cache lifetime

Agentic Strength

Leads its predecessor across BrowseComp, OSWorld 2.0 and Agents’ Last Exam, with major gains in browser and desktop automation
Luna scores only 41.3% on the 256K-1M long-context evaluation — for very long documents, prefer Terra / Sol or chunk the input.

​
Practical Use

​
Which Tier to Pick

  • gpt-5.6-sol: ceiling-bound agentic work — complex coding agents, desktop / browser automation, deep research
  • gpt-5.6-terra: the daily driver — GPT-5.5-class performance at half the price, the natural migration target for most production workloads
  • gpt-5.6-luna: high-volume traffic — classification, extraction, light chat, tool-calling glue

​
Code Example

import openai

client = openai.OpenAI(
    api_key="your-apiyi-key",
    base_url="https://api.apiyi.com/v1"
)

response = client.chat.completions.create(
    model="gpt-5.6-terra",  # or gpt-5.6-sol / gpt-5.6-luna
    messages=[
        {"role": "user", "content": "Summarize the differences between the three GPT-5.6 tiers"}
    ]
)
print(response.choices[0].message.content)
Existing GPT-5.5 code migrates by swapping the model field.

​
Pricing & Availability

​
Pricing (per 1M tokens, matching official rates)

ModelInputOutputTier
gpt-5.6-sol$5.00$30.00Flagship
gpt-5.6-terra$2.50$15.00Balanced
gpt-5.6-luna$1.00$6.00Lightweight

​
Two Supply Groups

GroupPricingNotes
Default official relay (default / svip)Official parityOfficial API resources; recharge bonus up to 20%, effectively about 17% off (1/1.2)
CodexReverseDefault 0.7x discountCodex reverse-engineered economy channel; availability tracks official Codex
Pick by need: the default official-relay group for production-grade stability, the CodexReverse group for cost-sensitive batch workloads. See the recharge promotions guide.

​
Summary & Recommendations

The real story of GPT-5.6 isn’t any single benchmark — it’s the tiered product structure. Terra delivers GPT-5.5-class performance at half the price and is the tier most worth migrating to first; Sol tops nearly every agentic benchmark (except SWE-Bench Pro, where Claude Fable 5 keeps a ~15-point lead); Luna gives high-concurrency workloads a new $1-input option. Run your existing workload through Terra first, and step up to Sol for agent-heavy scenarios.
Sources: the official OpenAI announcement at openai.com/index/gpt-5-6, the OpenAI Help Center, MarkTechPost, and VentureBeat; data retrieved July 10, 2026. All three models are live on APIYI — grab a key and start calling.