GPT-6 Sol, Luna and Opus 5.5: Enterprise AI Pricing Reset
GPT-6 Sol, Luna and Opus 5.5: Enterprise AI Pricing Reset
September 22, 2026 is the day enterprise AI costs reset. OpenAI and Anthropic each launched new models within hours of each other, cutting access to near-frontier AI by 40 to 50 percent compared to last week’s pricing. For enterprise teams budgeting AI infrastructure or scaling agentic workloads, the math changed today.
Here is what launched, what it costs, and how the two families compare.
OpenAI Expands GPT-6 With Sol and Luna
OpenAI’s official announcement confirms two new additions to the GPT-6 lineup: GPT-6 Sol and GPT-6 Luna. Both models were trained using similar methods to the flagship GPT-6 Astra, which launched September 3. The company says the improvements behind Astra’s state-of-the-art performance in factuality, coding, computer use, and alignment carry over to the new models.
GPT-6 Sol is designed for complex coding and agentic workflows. On OpenAI’s internal factuality evaluation (drawn from real user-reported mistakes), GPT-6 Sol makes about half as many mistakes as GPT-5.6 Sol, “reaching Astra-level reliability at much lower cost,” according to the company. The model is available today through the OpenAI API, ChatGPT Work, and Codex for Plus, Pro, Business, Enterprise, and Edu users.
GPT-6 Luna is built for high-volume, goal-oriented tasks: document summarization, information extraction, Q&A over structured data, and similar clerical automation. Free and Go users can access Luna in the desktop app. Enterprise administrators must enable the new models in workspace settings.
The headline pricing change: both models come in at 50% of the promotional cost of their GPT-5.6 predecessors. OpenAI attributed the reduction to infrastructure improvements in caching and inference efficiency.
Anthropic Releases Claude Opus 5.5
Hours earlier, Anthropic published Claude Opus 5.5 on its product announcement page. This is the first model Anthropic has shipped since CEO Dario Amodei’s September 12 essay calling for deliberate pacing at the AI frontier. The company leaned into that framing: Opus 5.5 is positioned not as a new capability frontier but as a safer, more affordable version of frontier-level performance.
Key claims from Anthropic:
- Opus 5.5 matches Claude Fable 5.1 on most coding and knowledge work benchmarks.
- It costs 40% less to run than Opus 5 ($4/$20 vs $5/$25 per million input/output tokens).
- It is faster to serve, reflecting lower compute requirements per token.
- It passed Anthropic’s most comprehensive automated behavioral audit, outperforming all prior models.
- External evaluators at Frontier Design and METR reviewed it before release.
- Attempted sandbox escapes are 85% lower than Opus 5, and every attempt was low severity and self-reported, according to The Verge’s reporting.
Claude Opus 5.5 is now available on the Claude API, Amazon Web Services Bedrock, Google Cloud, and Microsoft Foundry. Anthropic says Claude Sonnet 5.5 and Haiku 5.5 will follow “in the coming weeks” with similar improvements.
Model Pricing: September 22, 2026 Snapshot
The table below reflects published pricing as of today. All figures are per million tokens (MTok), standard (non-batch) rate.
| Model | Input $/MTok | Output $/MTok | vs Prior Gen | Primary Use |
|---|---|---|---|---|
| GPT-6 Astra | Not confirmed | Not confirmed | New tier | Flagship: most demanding tasks |
| GPT-6 Sol (new) | $2.00 | $10.00 | -50% vs 5.6 Sol | Coding, agentic workflows |
| GPT-5.6 Sol (replaced) | $4.00 | $20.00 | Reference | Coding (now superseded) |
| GPT-6 Luna (new) | $0.10 | $0.50 | -50% vs 5.6 Luna | High-volume, clerical tasks |
| GPT-5.6 Luna (replaced) | $0.20 | $1.20 | Reference | High-volume (now superseded) |
| Claude Fable 5.1 | $10.00 | $50.00 | Reference tier | Frontier agentic coding |
| Claude Opus 5.5 (new) | $4.00 | $20.00 | -40% vs Opus 5 | Long-running agentic + knowledge |
| Claude Opus 5 (replaced) | $5.00 | $25.00 | Reference | Replaced by 5.5 |
| Grok 4.7 | $2.00 | $6.00 | Same as 4.6 | Coding, knowledge work |
Sources: OpenAI API pricing, Anthropic platform docs, The New Stack.
What This Changes for Enterprise Teams
Budget recalculation is warranted today. If your team was running GPT-5.6 Sol for coding automation, switching to GPT-6 Sol immediately halves your per-token spend while delivering better factuality. The same holds for Luna-tier workloads. These are not gradual rollouts. Both models are live now in API and ChatGPT Work.
Anthropic’s positioning is deliberate. Releasing Opus 5.5 at Fable-level capability but Opus-tier pricing is a direct response to the price compression OpenAI began with GPT-6 Astra’s deployment. Enterprises that standardized on Opus 5 for reliability-sensitive tasks now get that reliability at a lower run rate without re-evaluating quality. As TechCrunch noted, it is the first Anthropic model to explicitly exceed Fable 5.1 on informal internal benchmarks while shipping at a cheaper price point.
The safety story matters for regulated industries. The 85% reduction in sandbox escape attempts, combined with external evaluation from METR and Frontier Design, gives compliance teams concrete documentation. For healthcare, financial services, and legal verticals (where AI hacking incidents have triggered scrutiny in recent weeks), Opus 5.5 arrives with a paper trail that prior models lacked.
Luna opens up new automation surface area. At $0.50 per million output tokens, GPT-6 Luna is below the threshold where most enterprise finance teams ask hard questions about AI spend per task. Workflows that were previously too high-volume to automate, including daily document processing, customer data extraction, and first-pass content review, become economical at Luna pricing. Enterprises can run Luna in ChatGPT Work today without standing up new infrastructure.
The dual-launch competitive dynamic. Both companies almost certainly knew the other was shipping today. The New Stack observed that OpenAI’s 50%-price announcement arrived after Opus 5.5 had already been published, making OpenAI’s comparison to Opus 5’s pricing outdated within hours. This is not a coincidence. It is the compressed release cadence the AI industry now operates at: price and capability moves happen in the same 12-hour window, and enterprise procurement teams need to track both simultaneously.
What Stays the Same
Neither launch replaces the flagship tier. Claude Fable 5.1 and GPT-6 Astra remain the top options for the most complex agentic tasks: multi-day code projects, legal research, complex financial modeling, and adversarial red-teaming. Today’s launches are designed to democratize access to near-frontier performance, not to retire the frontier itself.
Grok 4.7, released September 21, continues to offer competitive pricing at $2/$6 per million tokens with a strong electrical-engineering benchmark showing. For teams that do not require deep OpenAI or Anthropic ecosystem integration, Grok 4.7 remains a credible third option.
Enterprise Stack Implications
Teams running Enera’s agentic GTM systems should evaluate where Sol-class or Luna-class throughput can replace costlier Astra or Fable calls in non-critical pipeline branches. The enterprise AI pricing revolt documented in August framed the problem correctly: token costs were the gating factor on agentic scale. Today’s announcements remove a significant share of that constraint.
For teams evaluating long-running knowledge work tasks, Opus 5.5 is now a direct competitor to GPT-6 Sol on price (both around $2 to $4 per million input tokens) while bringing Anthropic’s compliance documentation and expanded safety evaluation. The GPT-6 Astra launch analysis from September 4 outlined why OpenAI’s GPT-6 generation would create downstream pricing pressure across the tier stack. That pressure arrived on schedule today.
Sources: Anthropic Opus 5.5 announcement, Anthropic platform model docs, TechCrunch: Anthropic Opus 5.5, The Verge: Opus 5.5 safety, 9to5Mac: GPT-6 Sol and Luna, TechCrunch: OpenAI GPT-6 Sol/Luna, The New Stack: GPT-6 pricing, OpenAI API docs, CNA