R4XNBlog
Visit r4xn.com

News

OpenAI’s GPT-6 Sol & Luna Face Anthropic’s Claude Opus 5.5: What Actually Changed on Sep 22

R4XN TECH5 min read92 views

Side-by-side pricing comparison of OpenAI GPT-6 Sol Luna and Anthropic Claude Opus 5.5 API rates

On the same day in September 2026, Anthropic launched Claude Opus 5.5 and OpenAI followed with GPT-6 Sol and Luna—cheaper mid-tier models with million-token contexts. Here is what was verified, how the prices stack up, and what builders should test this week.

Same-day dual launch: Opus 5.5 and GPT-6 Sol/Luna on 22 September 2026.

What happened

On 22 September 2026, Anthropic released Claude Opus 5.5, the first model in its Claude 5.5 family. Roughly an hour later, OpenAI expanded the GPT-6 line beyond its earlier flagship GPT-6 Astra with two efficiency-focused models: GPT-6 Sol and GPT-6 Luna.

Reputable coverage from TechCrunch and The Verge, plus primary posts from Anthropic and OpenAI’s developer community, confirm this was a real same-day product race—not a rumor cycle.

OpenAI positioning (vendor claims)

  • Sol and Luna build on GPT-6 Astra’s advances, aimed at faster and cheaper work at scale.
  • Sol targets complex coding and agentic workflows.
  • Luna targets high-volume clerical work: summarization, extraction, and quick Q&A.
  • API prices cut roughly 50% versus GPT-5.6 promotional pricing, attributed to caching and inference efficiency.
  • Rollout: ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu; both models in the API as gpt-6-sol and gpt-6-luna; Free and Go users can try Luna in the desktop app.

Anthropic positioning (vendor claims)

  • Opus 5.5 performs at roughly Claude Fable 5.1 level on most work while costing about 40% less than Opus 5 on typical workloads.
  • Stronger focus on long-running agentic coding, knowledge work, communication clarity, and alignment/safeguards.
  • Available day one via Claude’s platform and major clouds (AWS, Google Cloud, Microsoft).
  • Sonnet 5.5 and Haiku 5.5 are promised in the coming weeks.

Independent analyst Simon Willison framed the day as a price war below the top flagship tier (Astra / Fable), where list prices remain much higher.

List-price stack: Luna for volume, Sol vs Opus 5.5 in the mid-tier, flagships still far above.

Pricing comparison

Figures below are standard API list prices per 1 million tokens, from vendor documentation and official announcements. Cache, batch, and fast modes change the real bill.

Model Input / MTok Output / MTok Cached input Context Best for GPT-6 Luna $0.10 $0.50 $0.01 1.05M (128K max out) High-volume efficient tasks GPT-6 Sol $2.00 $10.00 $0.20 1.05M (128K max out) Coding and agentic mid-tier Claude Opus 5.5 $4.00 $20.00 $0.20 (cache reads) 1M (128K max out) Flagship Opus refresh Claude Opus 5 (prior) $5.00 $25.00 $0.50 (cache reads) — Comparison only GPT-6 Astra / Claude Fable 5.1 ~$10 ~$50 — — Top flagships (not the Sep 22 mid-tier drop) Extra pricing facts worth budgeting for:

  • OpenAI: Batch/Flex at 50% of Standard; Fast mode at 2×; prompts over 272K input tokens billed at higher rates for the full request.
  • Anthropic: Fast mode for Opus 5.5 at $8 / $40 (up to about 2.5× speed); Batch 50% off; 5-minute cache writes $5, 1-hour $8.
  • Anthropic also raised five-hour usage limits on several subscription plans.

Primary sources: OpenAI GPT-6 Sol docs, GPT-6 Luna docs, Anthropic Opus 5.5 announcement, TechCrunch coverage.

What changed for builders

  1. Mid-tier got dramatically cheaper on OpenAI. Sol at $2/$10 undercuts Opus 5.5’s $4/$20 on list price. Luna at $0.10/$0.50 sits near the bottom of hosted frontier pricing and is explicitly aimed at volume workloads.
  2. Anthropic’s Opus tier got cheaper and (per Anthropic) faster. Opus 5.5 cuts token prices 20% versus Opus 5 and claims more than 30% faster generation, with a steeper drop on cache reads—important for agent loops where most tokens are cached.
  3. Specs converged at the long-context ceiling. Sol, Luna, and Opus 5.5 all advertise roughly 1M context and 128K max output. Knowledge cutoffs differ (OpenAI Sol Apr 20, 2026; Luna May 18, 2026; Opus 5.5 June 2026 per Anthropic docs).
  4. Capability claims conflict—treat benches as marketing until you reproduce them. OpenAI highlights fewer factuality mistakes for Sol versus its predecessor. Anthropic publishes tables claiming Opus 5.5 leads on several agentic coding and knowledge-work benches. Artificial Analysis reported Sol/Luna capturing a large share of the cost-efficiency frontier, with a mix of gains and regressions versus GPT-5.6.
  5. Migration friction is asymmetric. Opus 5.5 ships with documented breaking changes for apps built on Opus 5. OpenAI’s Sol/Luna pitch emphasizes drop-in affordability and higher usage limits, with Free/Go Luna access on desktop.
  6. Cloud availability differs on day one. Anthropic stresses multi-cloud availability for Opus 5.5. OpenAI’s launch post centers on its own API, ChatGPT Work, and Codex—teams buying models only through a hyperscaler should confirm availability before rewriting production routes.

Practical routing: volume to Luna, everyday agents to Sol, hardest jobs to Opus 5.5—after your own evals.

What to try this week

  1. Run a fixed eval set (10–20 real tasks: code fix, doc extract, agent tool loop) on your current default, plus Sol, Luna, and Opus 5.5. Measure cost per successful task, not tokens alone.
  2. Route by workload: Luna for bulk summarization/extraction; Sol for everyday coding agents if you want OpenAI’s mid price; Opus 5.5 for the hardest multi-hour jobs if you already trust Claude.
  3. If you use Claude agents, read Anthropic’s migration notes before flipping production IDs to claude-opus-5-5.
  4. If you’re Free/Go on ChatGPT, try Luna on desktop and decide whether paid Work/Codex access is worth it for Sol.
  5. Hold a light wait decision on Anthropic’s mid/cheap tiers until Sonnet 5.5 and Haiku 5.5 land—especially if your app is price-sensitive at Haiku/Sonnet volumes.
  6. Don’t rewrite your whole stack this week. Pin model IDs, A/B in staging, and revisit after independent evals settle.

Bottom line

22 September 2026 was a verified dual launch: Claude Opus 5.5 cut Opus pricing and pushed Anthropic’s flagship-adjacent capability story, while GPT-6 Sol and Luna brought Astra-era features to sharply lower OpenAI mid-tier prices—with Luna especially aggressive on cost. For founders and SMEs, the actionable move is not “pick a winner,” but re-benchmark your real workloads, route cheap volume to Luna (or wait for Anthropic’s mid-tier refresh), and treat Sol versus Opus 5.5 as a cost-versus-capability choice you’ll only settle with your own logs.