Anthropic launches Opus 5.5 as OpenAI ships GPT-6 Sol and Luna

Why it matters

A cheaper GPT-6 tier would make OpenAI's newest model generation practical for coding agents and high-volume applications, but the circulated price and speed claims remain unconfirmed.

Anthropic and OpenAI turned Tuesday's model rumors into three official launches. Anthropic introduced Claude Opus 5.5, while OpenAI launched GPT-6 Sol and GPT-6 Luna for lower-cost coding, agent and high-volume workloads.

Anthropic said Opus 5.5 is the first model in its Claude 5.5 family. The company said it performs at the level of Claude Fable 5.1 for most tasks and costs 40% less to run than Opus 5. Anthropic did not provide absolute token prices in the announcement.

Tuesday's timeline

• Earlier Tuesday: BridgeMind AI said Opus 5.5 had appeared in Claude Code. Some users said the model remained unavailable, while another reported gaining access after installing Claude Code version 2.1.280, suggesting a limited or staggered rollout.
• 12:31 p.m. ET: Anthropic's Claude account officially introduced Opus 5.5 and identified it as the first release in the Claude 5.5 family.
• Later Tuesday: OpenAI officially introduced GPT-6 Sol and GPT-6 Luna, confirming both model names and making them available through ChatGPT Work, Codex and the API.
• At launch: GPT-6 Sol and Luna began rolling out to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex. Free and Go users received access to Luna through the desktop app. OpenAI said neither model was yet available in the regular Chat interface.

The earlier Anthropic appearance followed claims from Pankaj Kumar (@pankajkumar_dev), who said he had tested Opus 5.5 and expected it to launch alongside GPT-6 Sol. Kumar put Opus 5.5's price at $4 per million input tokens and $20 per million output tokens, and described it as competitive with Fable 5.1 in 3D and frontend work, though not token-efficient. Anthropic's announcement supports the broad Fable 5.1 performance comparison, but Kumar's pricing and workload-specific claims remain unconfirmed.

Kumar also predicted that GPT-6 Sol would offer similar capability and pricing. The launch confirmed the model but not his expected price. OpenAI prices GPT-6 Sol at $2 per million input tokens and $10 per million output tokens, half the promotional rates for GPT-5.6 Sol.

The OpenAI claims had spread through posts from AI users watching model identifiers in Microsoft Azure and Codex sessions. KC (@karanC_12) described GPT-6 Sol as a model priced at $2.50 per million input tokens and $15 per million output tokens, with roughly six times the speed of GPT-5.6 Sol. The post also cited an early test that produced about 28,000 tokens in three minutes.

The launch confirms that the model was real, but the circulated pricing was high. OpenAI's published $2/$10 rates are 20% below the rumored input price and about 33% below the rumored output price. OpenAI did not use the six-times-faster claim in its launch announcement.

Salio (@Mr_Salio) said a ChatGPT Pro user's Codex session routed 12 requests to a model identified as gpt-6-sol, and claimed OpenAI was also preparing GPT-6 Luna. Both identifiers proved accurate: the callable API model IDs are gpt-6-sol and gpt-6-luna.

OpenAI cuts prices across two GPT-6 tiers

GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens. Those rates are 50% below GPT-5.6 Sol's promotional $4/$20 pricing.

GPT-6 Luna costs $0.10 per million input tokens and $0.50 per million output tokens. GPT-5.6 Luna costs $0.20 and $1.20, respectively, making the new model 50% cheaper for input and about 58% cheaper for output.

Both models support reasoning settings from none through max, a 1.05 million-token context window and outputs of up to 128,000 tokens. OpenAI's documentation lists support for web search, file search, image generation, code execution, computer use and MCP connections through the Responses API.

The lower pricing gives developers a broader ladder beneath GPT-6 Astra. Sol is positioned for complex coding and agentic workflows, while Luna targets focused tasks that need to run cheaply at high volume. An agent can reserve Astra or Sol for consequential decisions while routing classification, extraction, tests and tightly specified tool calls to Luna.

The earlier throughput report of 28,000 tokens in three minutes works out to roughly 156 tokens per second. That remains an anecdotal measurement rather than a controlled comparison. The original posts provided no consistent baseline, workload, reasoning setting, context size or processing tier for the assertion that Sol was six times faster.

OpenAI sells capability per dollar

OpenAI framed the release around cost per completed task rather than raw benchmark leadership. The company said GPT-6 Sol at its xhigh reasoning setting scored 33.2% on AutomationBench, compared with 26.9% for Claude Opus 5 at max effort. OpenAI calculated Sol's cost at $0.27 per task, about 9% of the Claude model's cost under the tested configurations.

On DeepSWE v1.1, OpenAI reported a 68.8% score for Sol at max effort, 1.1 percentage points behind the result it cited for Claude Fable 5 at xhigh effort. OpenAI said Sol cost roughly 80% less per task in that comparison. Luna scored 66.6% at max effort, according to the company.

Those are vendor benchmarks using OpenAI's selected reasoning settings, prices and evaluation harness. They establish OpenAI's positioning for the models, not an independent finding about production performance or total workload cost.

OpenAI also said Sol produces about half as many factual errors as GPT-5.6 Sol on an internal evaluation built from conversations where users flagged mistakes. The company acknowledged that the test uses error-inducing conversations rather than a representative sample of ordinary ChatGPT traffic.

Cheaper context is part of the release

OpenAI paired the models with prompt-caching changes for agents and long-running conversations. Cached input reads receive a 90% discount, and developers can change reasoning effort or tool availability without automatically invalidating earlier cached context.

That matters for systems that repeatedly send instructions, files, conversation history and tool definitions to a model. OpenAI cited GitHub data showing that caching improvements reduced freshly processed prompt tokens by more than 50% across billions of requests, although the announcement did not isolate how much of that reduction came from Sol and Luna.

The release fills the price gap beneath GPT-6 Astra, which remains OpenAI's highest-capability model at $10 per million input tokens and $50 per million output tokens. Sol now provides an Astra-era model at one-fifth of those rates, while Luna pushes GPT-6 into workloads where inference volume can determine whether a product is economical to operate.

The day's evidence is now decisive on both sides: Anthropic has officially introduced Opus 5.5, and OpenAI has launched GPT-6 Sol and Luna with lower prices than the figures circulated before release.