GPT-6 Sol and Luna split complex work from high-volume tasks
OpenAI's two new reasoning models share API surfaces but target different workload shapes, making task routing and evaluation central to adoption.
Key takeaways
OpenAI released GPT-6 Sol and GPT-6 Luna in the API on September 22. Both reasoning models accept text and image input and return text through the Responses and Chat Completions APIs. OpenAI positions Sol for complex coding and agentic workflows and Luna for focused, high-volume work. For standard processing with prompts up to 272K input tokens, listed prices per million tokens are $2 input and $10 output for Sol, and $0.10 input and $0.50 output for Luna, with separate cached-input rates. Longer prompts and other processing tiers can differ. Teams should route by task difficulty, failure cost, latency target, and quality evidence rather than choosing one model solely from the headline token price.
Direct answer
OpenAI released GPT-6 Sol and GPT-6 Luna in the API on September 22. Both reasoning models accept text and image input and return text through the Responses and Chat Completions APIs. OpenAI positions Sol for complex coding and agentic workflows and Luna for focused, high-volume work. For standard processing with prompts up to 272K input tokens, listed prices per million tokens are $2 input and $10 output for Sol, and $0.10 input and $0.50 output for Luna, with separate cached-input rates. Longer prompts and other processing tiers can differ. Teams should route by task difficulty, failure cost, latency target, and quality evidence rather than choosing one model solely from the headline token price.
Verified facts
The OpenAI API changelog records both models as released on September 22 with text and image input through Responses and Chat Completions.
At standard processing for prompts up to 272K tokens, Sol is listed at $2 input, $0.20 cached input, and $10 output per million tokens; Luna is $0.10, $0.01, and $0.50.
OpenAI positions Sol for complex coding and agentic workflows and Luna for focused high-volume work; longer context and other processing tiers use different pricing.
What changed
- One API gains two workload-oriented model tiers
- Text and image inputs use familiar API surfaces
- The pricing gap expands the case for task routing
- Long-context requests require separate cost accounting
Impact for AI users
Teams can evaluate Luna for frequent classification, extraction, and formatting, and Sol for complex coding, longer reasoning chains, and multi-tool work. The final boundary must come from measured outcomes. A router should also identify high-consequence failures and escalate to a stronger model or a human when confidence is low, budgets are exceeded, or tools behave unexpectedly.
Operating checklist
- Label samples by difficulty and failure consequence
- Record quality, latency, tokens, and tool calls for each model
- Define stronger-model and human-escalation conditions
- Observe routing errors and spend under a small canary
AI frontier news and analysis, AI software and model tools, AI skill tutorials and validation methods, and AI account and permission guidance
FAQ
Can lower-priced Luna handle every task?
Price does not define the capability boundary. Validate representative tasks and add escalation for high-risk or low-confidence requests.
Do prompts above 272K use the same rate?
No. OpenAI states that longer-context pricing differs, so it must be checked and budgeted separately.
Summary
Sol and Luna make model choice a task-level decision. A clear evaluation set, escalation policy, and cost monitoring matter more than assigning one model to every request.
This AI-assisted article is checked by ENHE AI automation for official sources, bilingual fields, media rights, page safety, and historical duplication before publication.
What this means for everyday users
Teams can evaluate Luna for frequent classification, extraction, and formatting, and Sol for complex coding, longer reasoning chains, and multi-tool work. The final boundary must come from measured outcomes. A router should also identify high-consequence failures and escalate to a stronger model or a human when confidence is low, budgets are exceeded, or tools behave unexpectedly.
Tools you may use
Related tutorials
Related Tools And Tutorials
Use the following ENHE AI sections to continue from the news signal into tool selection, account-service guidance, or practical learning.
Related reading
GitHub Copilot Model Retirement on November 16: A Migration Checklist
GitHub announced on October 15, 2026 that three models will be removed from GitHub Copilot experiences on November 16: Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra. The change covers Copilot Chat, inline edits, ask and agent modes, and code completions. GitHub recommends Claude Opus 5.5 for both Claude models and either GPT-6.1 Sol or GPT-6 Luna for GPT-5.6 Terra. Under default model enablement, suggested alternatives are enabled automatically for Copilot Business and Enterprise unless an administrator disabled the global default or the specific model. Claude Opus 4.8 remains available to individual annual subscribers as an exception. This guide provides a durable migration checklist for inventorying model references, checking policies, testing prompts and integrations, comparing quality and cost, communicating the switch, and keeping a rollback path before the deadline.
GPT-6.1 Sol brings near-Astra capability to a lower-cost production tier
OpenAI made GPT-6.1 Sol available on September 29 as a lower-cost model for complex coding, computer use, and professional work with performance positioned near GPT-6 Astra. The official model page lists a 1.05-million-token context window, a 128,000-token output limit, five supported reasoning-effort levels, and Responses API tool calling. Standard API pricing is listed at $2 per million input tokens and $10 per million output tokens. A same-day safety addendum classifies cybersecurity capability as Critical and biological and chemical capability as High, with the Astra safeguard stack applied. Teams should route real tasks through a controlled comparison of quality, latency, total token use, tool success, refusals, and data-residency requirements before changing a production default.
Amazon Bedrock and OpenCode make multi-model coding agents auditable
AWS described on September 23 how to use open-weight models in OpenCode through Amazon Bedrock. OpenCode is a terminal-native Go agent that supports more than 75 model providers; the example uses Kimi K3, GPT-OSS 120B, and NVIDIA Nemotron 3 Super 120B while keeping inference in the customer’s AWS account. IAM, CloudTrail, PrivateLink, encryption, and model routing provide controls that fit existing delivery practices. As a durable task, teams should pin model and permission versions, isolate the workspace, log tool calls and cost, and place real-service verification after local tests before allowing an agent patch into release. Teams should verify the vendor claims against their own workloads, controls, and operating context before making a production decision.
AWS benchmarks OpenAI models by cost per correct outcome instead of token price alone
AWS published an open-source benchmarking approach on September 11 for comparing GPT-5.6 Luna, Terra, and Sol on Amazon Bedrock with GPT-5.4 mini and nano on the OpenAI API. Instead of ranking models by dollars per million tokens, the harness measures cost per correct answer, cost per passing multi-turn research result, and cost per rubric-accepted professional deliverable. It uses one Responses API client and records timestamped result files, frozen grading prompts, and chart inputs. The post also documents important limits: sample sizes range from 48 to 198 items, Bedrock models ran with reasoning disabled, and API baselines used their defaults. AWS describes this as a practical configuration comparison rather than an estimate of intrinsic capability. Teams should rerun the method on 50 to 100 of their own tasks and include failures, retries, turns, latency, review, and rework in the decision.
OpenAI launches GPT-6 Astra with stronger computer use and explicit enterprise enablement
OpenAI introduced GPT-6 Astra on September 3 with major upgrades in computer use, browsing, software engineering, science, and professional work. The model is rolling out in phases to ChatGPT plans and is also available through the OpenAI API, Microsoft Azure, and AWS Bedrock. Enterprise access is off by default at launch and must be enabled by an administrator. OpenAI lists standard API pricing of $10 per million input tokens and $50 per million output tokens, with separate cache rates. It also classifies Astra at the Critical cybersecurity capability threshold and applies stronger safeguards. Teams should treat the reported benchmarks as vendor evidence, then run their own task, permission, latency, cost, and rollback tests before broad deployment.
OpenAI Previews GPT-5.6 Sol Ultrafast at Up to 14x Speed
OpenAI previewed GPT-5.6 Sol Ultrafast on August 13, 2026, saying the Cerebras-powered service can run up to fourteen times faster than the standard version and reach up to roughly 750 output tokens per second. The company is starting with selected API customers; it did not announce universal ChatGPT access, one public price, or a general availability date. Developers should confirm invitation status and region, then compare the same long-context, structured-output, or agent task against the standard model. Record time to first token, total latency, accepted quality, tool behavior, cost, limits, and retries. Peak throughput alone is not enough evidence to replace a production model.