- Joined
- Mar 29, 2014
- Messages
- 2,349
- Reaction score
- 884
I've just learnt at
that the OpenAI is changing how it names and positions its models.
// super interesting, especially if you start exploring agents (like I do right now with Hermes Agents).
.. ok, let' s dive into this (with help of chatgpt itself):
>> Instead of introducing another sequence of temporary labels such as “Turbo”, “Preview”, “Instant” or “Pro”, the GPT-5.6 generation is being divided into three capability tiers: Sol, Terra and Luna.
The model number will identify the generation, while the new names are intended to represent more permanent positions within the family.
In simple terms:
The structure looks similar to Anthropic’s Haiku, Sonnet and Opus approach. Users may finally be able to understand where a model belongs without decoding a growing collection of version numbers and product suffixes.
Sol: the model for difficult work
GPT-5.6 Sol is being positioned as OpenAI’s most capable model.
It introduces a new “max” reasoning effort for tasks where the model needs more time to think. OpenAI is also introducing an “ultra” mode that can use sub-agents to work on complex problems beyond what one model instance might handle efficiently.
That makes Sol the obvious candidate for demanding workflows such as:
For Hermes Agent users, Sol could become the model reserved for the jobs where a cheaper model begins looping, misses dependencies or needs too much human correction.
It is also the most expensive model in the family.
Terra: likely to become the daily workhorse
Terra sits in the middle.
OpenAI describes it as a balanced model for everyday work and says its performance is competitive with GPT-5.5 while costing approximately half as much.
That combination could make Terra more important in practice than Sol.
Most developers and businesses do not need the strongest possible model for every request. They need a model that is reliable enough for coding, writing, analysis, research and automation without turning every workflow into a large API bill.
Terra therefore looks like the natural default for agent systems. Sol can be called only when a task becomes difficult enough to justify the additional cost.
Luna: built for speed and volume
Luna is the fastest and least expensive GPT-5.6 tier.
It is aimed at tasks where throughput matters more than maximum reasoning ability, including:
For an AI agent, Luna may be useful for the hundreds or thousands of small operations that happen around the main task.
A system could use Luna to inspect files, classify requests and prepare data, Terra to complete the normal work, and Sol only when deeper reasoning is required.
That is where these three tiers become more interesting than a simple model comparison. They appear designed for routing.
GPT-5.6 API pricing
OpenAI has listed the following prices per one million tokens:
The difference between Luna and Sol is substantial, particularly for output tokens.
Using Sol for every step of an automated workflow would be difficult to justify when many of those steps could be completed by Terra or Luna.
This may encourage developers to stop thinking about “the best model” and start designing systems that choose the right model for each individual job.
What this could mean for domain professionals
The three-tier structure maps surprisingly well to domain industry workflows.
Luna could process large domain lists, classify names, extract keywords, clean marketplace data or produce initial summaries.
Terra could handle daily research, sales descriptions, outreach preparation, portfolio analysis and development work.
Sol could be reserved for complex valuation research, legal or policy analysis, software architecture, large acquisition studies and multi-agent investigations.
The real opportunity may not be gaining access to a smarter chatbot. It may be building tools that know when intelligence is worth paying for.
A more stable naming system
OpenAI says Sol, Terra and Luna are intended to be durable capability tiers that can improve on their own schedules.
This is an important distinction.
The company could eventually update Terra without replacing it with another confusing product name. Users would continue to understand that Terra represents the balanced tier, even as the underlying model improves.
Stable names should also make configuration easier for AI applications, model routers and agent platforms such as Hermes Agent.
The original discussion appeared in the Hermes Agent community:
OpenAI’s GPT-5.6 announcement: https://openai.com/index/previewing-gpt-5-6-sol/
The old habit was to select one model and send almost everything to it.
GPT-5.6 appears to be pushing developers toward a different architecture: Luna handles the volume, Terra runs the business, and Sol is brought in when the problem becomes expensive enough to deserve it.
// super interesting, especially if you start exploring agents (like I do right now with Hermes Agents).
.. ok, let' s dive into this (with help of chatgpt itself):
>> Instead of introducing another sequence of temporary labels such as “Turbo”, “Preview”, “Instant” or “Pro”, the GPT-5.6 generation is being divided into three capability tiers: Sol, Terra and Luna.
The model number will identify the generation, while the new names are intended to represent more permanent positions within the family.
In simple terms:
- Sol is the flagship model for difficult reasoning, coding, research and agentic work.
- Terra is the balanced model for everyday professional use.
- Luna is the faster and less expensive model for high-volume tasks.
The structure looks similar to Anthropic’s Haiku, Sonnet and Opus approach. Users may finally be able to understand where a model belongs without decoding a growing collection of version numbers and product suffixes.
Sol: the model for difficult work
GPT-5.6 Sol is being positioned as OpenAI’s most capable model.
It introduces a new “max” reasoning effort for tasks where the model needs more time to think. OpenAI is also introducing an “ultra” mode that can use sub-agents to work on complex problems beyond what one model instance might handle efficiently.
That makes Sol the obvious candidate for demanding workflows such as:
- Software architecture and deep debugging
- Long-running coding tasks
- Agent design and orchestration
- Cybersecurity and vulnerability research
- Scientific and technical research
- Complex business analysis
For Hermes Agent users, Sol could become the model reserved for the jobs where a cheaper model begins looping, misses dependencies or needs too much human correction.
It is also the most expensive model in the family.
Terra: likely to become the daily workhorse
Terra sits in the middle.
OpenAI describes it as a balanced model for everyday work and says its performance is competitive with GPT-5.5 while costing approximately half as much.
That combination could make Terra more important in practice than Sol.
Most developers and businesses do not need the strongest possible model for every request. They need a model that is reliable enough for coding, writing, analysis, research and automation without turning every workflow into a large API bill.
Terra therefore looks like the natural default for agent systems. Sol can be called only when a task becomes difficult enough to justify the additional cost.
Luna: built for speed and volume
Luna is the fastest and least expensive GPT-5.6 tier.
It is aimed at tasks where throughput matters more than maximum reasoning ability, including:
- Summaries
- Classification
- Transcript and data cleanup
- Bulk translation
- Content extraction
- RAG preprocessing
- Simple customer support
- Background automation
For an AI agent, Luna may be useful for the hundreds or thousands of small operations that happen around the main task.
A system could use Luna to inspect files, classify requests and prepare data, Terra to complete the normal work, and Sol only when deeper reasoning is required.
That is where these three tiers become more interesting than a simple model comparison. They appear designed for routing.
GPT-5.6 API pricing
OpenAI has listed the following prices per one million tokens:
| Model | Input | Output |
|---|---|---|
| GPT-5.6 Sol | $5.00 | $30.00 |
| GPT-5.6 Terra | $2.50 | $15.00 |
| GPT-5.6 Luna | $1.00 | $6.00 |
The difference between Luna and Sol is substantial, particularly for output tokens.
Using Sol for every step of an automated workflow would be difficult to justify when many of those steps could be completed by Terra or Luna.
This may encourage developers to stop thinking about “the best model” and start designing systems that choose the right model for each individual job.
What this could mean for domain professionals
The three-tier structure maps surprisingly well to domain industry workflows.
Luna could process large domain lists, classify names, extract keywords, clean marketplace data or produce initial summaries.
Terra could handle daily research, sales descriptions, outreach preparation, portfolio analysis and development work.
Sol could be reserved for complex valuation research, legal or policy analysis, software architecture, large acquisition studies and multi-agent investigations.
The real opportunity may not be gaining access to a smarter chatbot. It may be building tools that know when intelligence is worth paying for.
A more stable naming system
OpenAI says Sol, Terra and Luna are intended to be durable capability tiers that can improve on their own schedules.
This is an important distinction.
The company could eventually update Terra without replacing it with another confusing product name. Users would continue to understand that Terra represents the balanced tier, even as the underlying model improves.
Stable names should also make configuration easier for AI applications, model routers and agent platforms such as Hermes Agent.
The original discussion appeared in the Hermes Agent community:
OpenAI’s GPT-5.6 announcement: https://openai.com/index/previewing-gpt-5-6-sol/
The old habit was to select one model and send almost everything to it.
GPT-5.6 appears to be pushing developers toward a different architecture: Luna handles the volume, Terra runs the business, and Sol is brought in when the problem becomes expensive enough to deserve it.
