OpenAI launches GPT-6 Sol and Luna with 50% API price cuts, intensifying AI cost war
On September 22-23, OpenAI released GPT-6 Sol and GPT-6 Luna, two new AI models with API prices permanently reduced by 50% compared to GPT-5.6 promotional pricing. GPT-6 Sol costs $2 per million input tokens and $10 output; GPT-6 Luna costs $0.10 input and $0.50 output. The models are available to Plus, Pro, Business, and Enterprise users via ChatGPT Work and Codex, but not yet in standard chat. Anthropic simultaneously launched Claude Opus 5.5 with similar cost reductions, escalating the AI token price war.
IllustrationEditorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Common ground
- OpenAI's 50% price cuts on GPT-6 Sol and Luna are a direct response to competitive pressure, not altruism.
- Chinese models like Xiaomi's MiMo-V2.6-Pro offer significant cost advantages for simple, high-volume tasks like translation and customer service.
- The price war is reshaping global AI access, especially for developing nations with limited budgets.
- Benchmark gaps exist between Western and Chinese models on complex tasks like coding and multi-step reasoning.
Points of contention
- Neutral Agent sees the price war as driven by inference efficiency gains, while Eastern Agent sees it as a forced reaction to Chinese competition.
- Eastern Agent argues Chinese models offer genuine cost advantages for most real-world tasks, while Neutral Agent claims the gap narrows on high-value workloads.
- Neutral Agent views Chinese data governance as opaque and risky, while Eastern Agent sees it as a transparent alternative to US surveillance.
- Eastern Agent believes the multipolar world is emerging, while Neutral Agent argues Western firms still lead on efficiency and profitability.
Blind spots
- Both sides overlook the role of specialized fine-tuned models in reducing reliance on general-purpose APIs for translation and customer service.
- The debate ignores how smaller AI firms in other regions (e.g., Europe, India) might affect the global price war.
- Neither agent fully addresses the long-term sustainability of state-backed Chinese AI funding versus Western venture capital models.
WorldAttention’s read
The debate reveals a fundamental split: Neutral Agent argues OpenAI's price cuts stem from genuine engineering advances in inference efficiency, making them a strategic leader, while Eastern Agent contends they are a defensive reaction to Chinese competition that is democratizing AI access for the Global South. Both agree that Chinese models offer cost advantages for simple tasks, but disagree on whether benchmark gaps on complex work matter for most users. The sovereignty argument remains unresolved, with Neutral Agent warning of opaque Chinese surveillance and Eastern Agent highlighting US extraterritorial data risks. Ultimately, the price war signals a shift toward more affordable AI, but whether it leads to a multipolar future or reinforces Western dominance depends on whether Chinese firms can sustain profitability and close performance gaps on high-value tasks.
Reporting timeline
The Intelligence Index vs Cost per Task Pareto frontier shifted this week with the releases of MiMo-V2.6-Pro, Claude Opus 5.5, GPT-6 Luna, and GPT-6 Sol Together they have established eleven new points on the Pareto frontier
This week, the Intelligence Index vs Cost per Task Pareto frontier shifted with the releases of MiMo-V2.6-Pro, Claude Opus 5.5, GPT-6 Luna, and GPT-6 Sol, establishing eleven new points on the frontier driven by different reasoning efforts. GPT-6 Luna (max) scores 37 at $0.068 per task, MiMo-V2.6-Pro scores 46 at $0.13, GPT-6 Sol (max) scores 48 at $1.06, and Claude Opus 5.5 (max with fallback) is the new highest-scoring model at 58 at $5.98. The five new points from GPT-6 Luna, one each from MiMo-V2.6-Pro and GPT-6 Sol, and four from Claude Opus 5.5 represent a significant update to the cost-efficiency landscape of frontier AI models.
OpenAI and Anthropic Launch New Models with Price Cuts, Competing on Cost-Efficiency
On September 22, Anthropic released Claude Opus 5.5 and OpenAI launched GPT-6 Sol and GPT-6 Luna, all emphasizing cost-effectiveness. OpenAI's GPT-6 Sol is priced at $2 per million input tokens and $10 per million output tokens, half the cost of its predecessor GPT-5.6 Sol. GPT-6 Luna costs $0.10 input and $0.50 output, with output prices down 58% from the prior model. OpenAI attributes the reductions to improved caching and inference efficiency. In benchmarks, GPT-6 Sol achieved a 33.2% score in AutomationBench at $0.27 per task, while Claude Opus 5 cost 11.1 times more for a 26.9% score. Anthropic's Opus 5.5 is priced at $4 input and $20 output per million tokens, 20% lower than Opus 5, with cache reads down 60%. Anthropic claims overall costs are about 40% lower than Opus 5 with 30% faster output. In Terminal-Bench 4.0, Opus 5.5 scored 66.4%, outperforming Fable 5.1 (55.8%) and Opus 5 (52.3%). Third-party evaluator Artificial Analysis rated Opus 5.5 highest in reasoning intensity at 58, versus GPT-6 Sol's 48 and GPT-6 Luna's 37. Separately, Meta's AI assistant Muse topped the U.S. App Store free chart on September 18, with over 2.5 million downloads by September 21. In China, Xiaomi released and open-sourced the MiMo-V2.6 series on September 22, maintaining previous API prices while improving intelligence, with the Pro model priced at 0.025 yuan per million input tokens (cache hit) and 6 yuan output, claimed to be the cheapest among domestic flagship models at 1/20 to 1/60 the price of overseas equivalents.
Read sourceOpenAI slashes GPT-6 API prices by half, challenging DeepSeek's low-cost advantage
OpenAI released two new models, GPT-6 Sol and GPT-6 Luna, on September 22, cutting API prices by 50% compared to GPT-5.6 promotional pricing. GPT-6 Luna's input price is as low as $0.10 per million tokens, directly undercutting DeepSeek's pricing in some scenarios. OpenAI attributed the price reduction to improved caching and inference efficiency, not subsidies. The move targets DeepSeek's price moat as AI competition shifts from capability to cost. Anthropic also launched Claude Opus 5.5 on the same day, emphasizing lower operating costs. GPT-6 Sol targets complex professional tasks, while Luna focuses on high-frequency, low-cost scenarios. The models are available via API and in ChatGPT Work and Codex for paid users, with free users able to test Luna on desktop. The rapid product iteration contrasts with industry calls from Anthropic and OpenAI executives to slow frontier AI development for safety reasons.
Read sourceShow 17 older updatesHide older updates
OpenAI cuts GPT-6 API prices by half, challenging DeepSeek's cost advantage
On September 22, OpenAI launched two new models, GPT-6 Sol and GPT-6 Luna, cutting their API prices by 50% from GPT-5.6 promotional pricing. GPT-6 Luna's input price is as low as $0.10 per million tokens, directly undercutting DeepSeek's low-cost offerings. OpenAI attributes the price reduction to improved caching and inference efficiency, passing savings to users. The move signals a shift in AI competition from capability to cost, as Anthropic also released Claude Opus 5.5 with lower operating costs on the same day. GPT-6 Sol targets complex tasks like coding and agent workflows, while Luna focuses on high-frequency, low-cost scenarios. The models are available to Plus, Pro, Business, Enterprise, and Edu users via ChatGPT Work and Codex, with free users able to test Luna on desktop. The article notes that industry leaders are simultaneously discussing slowing AI development while accelerating product releases and cost reductions.
Read sourceOpenAI Releases GPT-6 Sol and Luna Models, Permanently Cuts API Prices by 50%
On September 22, OpenAI officially launched two new models in its GPT-6 series: GPT-6 Sol and GPT-6 Luna. The API prices for both models have been permanently reduced by 50% compared to the previous promotional pricing for GPT-5.6. This release completes OpenAI's GPT-6 product matrix, which now includes the flagship GPT-6 Astra, the mid-range GPT-6 Sol, and the lightweight GPT-6 Luna. Specific pricing for GPT-6 Sol is set at $2 per million input tokens (down from $4) and $10 per million output tokens (down from $20). GPT-6 Luna is priced at $0.1 per million input tokens (down from $0.2) and $0.5 per million output tokens (down from $1.2). OpenAI stated that both new models offer improvements in intelligence, coding, and computer use over their GPT-5.6 counterparts, and that the permanent price reduction will make more application scenarios economically viable.
Read sourceOpenAI and Anthropic Cut Prices Up to 50% on New AI Models, Shifting Focus to Cost Efficiency
On September 22, OpenAI and Anthropic released new AI models within 90 minutes of each other, both emphasizing reduced inference costs and improved speed over raw performance gains. OpenAI launched GPT-6 Sol and GPT-6 Luna, which inherit capabilities from GPT-6 Astra and offer API prices permanently reduced by 50% compared to GPT-5.6 promotional pricing. Sol targets complex coding and agent workflows, while Luna is for cost-sensitive, high-volume tasks. Anthropic released Claude Opus 5.5, which matches Claude Fable 5.1 on most tasks but costs 40% less to run than Opus 5, with input/output token prices down 20% and cache read prices down 60%. The companies still pursue performance improvements, with Opus 5.5 exceeding Opus 5 and Fable 5.1 on some benchmarks and even surpassing GPT-6 Astra on certain tests. However, developer feedback was mixed; AI coding community founder Wes Winder questioned GPT-6 Sol's performance on the DeepSWE benchmark, prompting OpenAI product lead Thibault Sottiaux to defend the model's overall experience and efficiency. The debate reflects a broader industry shift from competing solely on model capability to competing on cost, speed, and scalable deployment, driven partly by Chinese open-source models gaining market share with better value.
Read sourceOpenAI Launches GPT-6 Sol and Luna, Sparking AI Price War with Deep Discounts
OpenAI released GPT-6 Sol and Luna, two new AI models priced significantly lower than its flagship Astra and competitor Anthropic's Opus 5.5. GPT-6 Sol costs $2 per million input tokens and $10 output, half the price of Opus 5.5, while Luna costs just $0.1 input and $0.5 output, about 1% of Astra's price. Both models maintain 105K token context and 128K output, supporting text and image inputs with tools like web search and code interpreter. Performance benchmarks show Sol achieving 33.2% on AutomationBench at $0.27 per task, close to Opus 5.5's 40% at $1.28. On DeepSWE v1.1, Sol scored 68.8% and Luna 66.6%, with costs 80-96% lower than competitors. The article notes that Chinese models like DeepSeek V4.1 Flash and Xiaomi's MiMo-V2.6-Pro offer competitive pricing, with MiMo-V2.6-Pro scoring 46 on Intelligence Index at $0.13 per task versus Sol's $1.06. The analysis concludes the AI model market is bifurcating into high-performance and low-cost strategies, with middle-ground models becoming unsustainable.
Read sourceGPT-6 Sol beats flagship Astra by 169 Elo at one-eighth cost and three times speed
A post from the account 'Whats_AI' reports that OpenAI's cheaper model, GPT-6 Sol, outperforms its flagship GPT-6 Astra on a writing benchmark for YouTube scripts. At the deepest effort setting, Sol scores 2182 Elo compared to Astra's 2013, a 169-point advantage. Sol costs $0.11 per script versus Astra's $0.85, and completes the task in 171 seconds versus Astra's 526 seconds. The post notes that every tested effort setting of Sol beats every setting of Astra, and that Astra's default setting (1843 Elo) falls below where GPT-5.6 Sol was. The author cautions that Astra is built for agentic and engineering work, and that the benchmark only measures prose quality, advising users to test the cheaper model on their own tasks before assuming the expensive one is better.
OpenAI's GPT-6 Sol and Luna halve cost with similar Intelligence Index scores
OpenAI has released two new models, GPT-6 Sol and Luna, which significantly improve cost efficiency. According to an analysis by Artificial Analysis, the models halve the cost of their predecessors while achieving similar scores on the Intelligence Index. The cost improvements are driven by approximately a 50% reduction in token pricing. The Artificial Analysis Intelligence Index chart illustrates the trade-off between intelligence and cost per task, showing that GPT-6 Sol and Luna push the cost efficiency frontier for OpenAI. This development marks a notable advancement in making high-performance AI models more affordable for users.
Read sourceGPT-6 Sol beats Astra on benchmark, ranks 6th of 94 models at $0.11 per script
In a post on X, the account Whats_AI announced that OpenAI's GPT-6 Sol outperformed Astra on its internal benchmark for YouTube script drafting. The model was tested on 10 scripts with five drafts each, scored blindly by three AI judges. GPT-6 Sol ranked 6th out of 94 models at a cost of $0.11 per script. The post highlights three key findings: the model's effort dial consistently improves results, with scores rising from 2058 at default to 2182 at max effort, a 124 Elo spread for roughly double the price; it beats both GPT-5.6 Sol and Astra while costing a third as much as GPT-5.6 Sol and significantly less than Astra; and it outperforms GPT-6 Astra by a wide margin. The author recommends users of GPT-5.6 Sol or Astra to upgrade to GPT-6 Sol at max effort, calling it a straight upgrade on both quality and price.
Read sourceOpenAI launches GPT-6 Sol and Luna with 50% price cut, expands model family
On September 22, OpenAI announced two new GPT-6 models, Sol and Luna, expanding its GPT-6 family less than three weeks after the flagship Astra launch. The models are priced 50% lower than GPT-5.6 promotional rates, attributed to improved caching and inference efficiency. Sol and Luna inherit Astra's technical advances but are optimized for speed and cost, targeting high-frequency coding, agent workflows, and enterprise automation. They are available immediately to Plus, Pro, Business, Enterprise, and Edu users via ChatGPT Work and Codex, as well as API; free and Go users can access Luna on desktop. The release signals a strategic shift from competing solely on capability to scaling cost-effective AI deployment, as rival Anthropic also launched a cheaper Claude Opus 5.5 model on the same day.
Read sourceGPT-6 Luna is half the price of 5.6 Luna, which was already an astonishingly cheap model given how capable it is Luna is my favorite model for building product features thanks to its cost (and speed)
In a post on X, developer Simon Willison states that the AI model GPT-6 Luna is priced at half the cost of GPT-5.6 Luna, which he describes as already being an astonishingly cheap model given its capabilities. Willison adds that Luna is his favorite model for building product features due to its cost and speed. The post presents a direct price comparison between two versions of the Luna model, highlighting a significant cost reduction for the newer iteration.
Read sourceOpenAI GPT-6 Sol and Luna models not yet available in Chat for Plus users
According to a post on X by user Mr_Salio, OpenAI's GPT-6 Sol and Luna models are not yet available in the standard Chat interface for Plus subscribers. The post cites OpenAI's website, which states: 'These models are not yet available in Chat.' However, the models are currently accessible through ChatGPT Work and Codex for Plus users. The author expresses frustration, questioning why OpenAI treats Plus users differently by not including the new models in the normal model picker. This development indicates a tiered rollout strategy for the latest GPT-6 models, limiting direct chat access for standard Plus subscribers while offering it through specialized work and coding platforms.
Read sourceGPT-6 Sol and Luna released with lower pricing and competitive benchmark scores against Opus 5
A post on X by pankajkumar_dev announces the official release of GPT-6 Sol and Luna, two new AI models. GPT-6 Sol is priced at $2 per million input tokens and $10 per million output tokens, while Luna is priced at $0.10 and $0.50 per million tokens, respectively, making it 50% cheaper than GPT-5.6 pricing. Benchmark results are provided: on AutomationBench, Sol scores 33.2% at $0.27 per task versus Opus 5's 26.9% at $3.00 per task. On DeepSWE, Sol achieves 68.8% compared to Fable 5's 69.9%, at around 80% lower cost per task. On OSWorld 2.0, Sol scores 60.5% versus Opus 5's 60.3%, also at roughly 80% lower cost. The post notes that the benchmarks used Fable 5, not Fable 5.1, and that Sol is not as good overall as Opus 5.5, which is why pricing was reduced. The author concludes it is a good upgrade for the price.
Read sourceOpenAI unveils GPT-6 Sol and Luna, claims up to 93% cheaper task costs than Claude Opus 5
OpenAI has announced the release of GPT-6 Sol and Luna, two new AI models. The company claims that these models achieve up to 93% lower task costs compared to Anthropic's Claude Opus 5 in certain coding evaluations. This announcement represents a significant competitive move in the AI model market, focusing on cost efficiency for coding tasks. The specific conditions and benchmarks for the claimed cost reduction were not detailed in the announcement.
Read sourceGPT-6 Sol and Luna launch with 50% price cut, Luna drops to $0.10 per million input tokens
The AI model provider has released GPT-6 Sol and Luna, claiming they are both better and 50% cheaper than the previous GPT-5.6 model. The new Luna model is priced at $0.10 per million input tokens and $0.50 per million output tokens. This price reduction follows an 80% price cut to Luna made at the end of July, which brought output pricing down from $6 to $0.50 per million tokens within two months. The announcement highlights a significant and rapid decrease in the cost of accessing advanced AI models.
Read sourceOpenAI's GPT-6 Sol and Luna deliver major improvements and half the price per token over 5.6-family predecessors
In a post on X, Sam Altman announced the release of GPT-6 Sol and Luna, stating they are big improvements on intelligence, alignment, work output, coding, computer use, and more over their 5.6-family predecessors. He also stated they are half the price per token, and even less per task. The post attributes these claims directly to the source and presents them as a factual announcement of a new product release.
Read sourceGPT-6 Sol and Luna released with significant improvements and 50% API price cut
In a post on X, user thsottiaux announced the release of GPT-6 Sol and Luna, describing them as a very significant improvement across the board, particularly in writing and overall quality. The post also states a permanent 50% reduction in the API price, making both models viable for new use cases and extending usage for subscribers. Additionally, a banked reset is being loaded into all accounts of Plus, Pro, and Business users. The announcement includes a link to further information.
Read sourceOpenAI launches GPT-6 Sol and GPT-6 Luna on OpenRouter at half the price of predecessors
OpenAI has released two new AI models, GPT-6 Sol and GPT-6 Luna, now available on the OpenRouter platform. The models are priced at half the cost of their GPT-5.6 predecessors. GPT-6 Sol is priced at $2 per million input tokens and $10 per million output tokens, while GPT-6 Luna is priced at $0.10 per million input tokens and $0.50 per million output tokens. According to the announcement, both models outperform their predecessors on the AutomationBench benchmark, achieving the best scores at a fraction of the cost per task. The post indicates that each model serves a different purpose, with further details promised in a thread. This release represents a significant price reduction and performance improvement in OpenAI's model lineup.
Read sourceOpenAI's GPT-6 Sol and Luna halve costs versus GPT-5.6, with mixed benchmark results
OpenAI has launched GPT-6 Sol and Luna, two new AI models that halve the cost per task compared to their GPT-5.6 predecessors. Pricing drops from $4/$20 to $2/$10 per million input/output tokens for Sol, and from $0.20/$1.20 to $0.10/$0.50 for Luna, with the same cache discounts. According to an analysis by Artificial Analysis, GPT-6 Sol (max) costs $1.06 per task on the Intelligence Index, about 50% less than GPT-5.6 Sol (max) at $1.99, while GPT-6 Luna (max) costs $0.07 per task, about 60% less. In the Coding Agent Index, Sol improved by 2 points to 57, while Luna regressed by 2 points to 41. Both models show significant reductions in hallucination rates on the AA-Omniscience benchmark: Sol dropped from 92% to 60% and Luna from 93% to 77%, though Sol achieved this by declining to answer more questions. Performance was mixed across other evaluations, with improvements in AutomationBench-AA and Terminal-Bench 4.0, but regressions in GDPval-AA v2.1 and AA-Briefcase v1.1, attributed to reduced presentation quality.
Read source