Anthropic and OpenAI Launch Cheaper AI Models, Intensifying Price War
On September 22-23, 2025, Anthropic released Claude Opus 5.5, claiming Fable 5.1-level performance at 40% lower cost than Opus 5, while OpenAI launched GPT-6 Sol and GPT-6 Luna with API prices 50% lower than predecessors. Both companies emphasized cost efficiency and safety improvements, with Anthropic planning additional 5.5 series models and OpenAI positioning Sol and Luna as affordable alternatives to its flagship Astra model.
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
Reporting timeline
There’s basically unanimous praise for Opus 5.5, but I think people *might* be overlooking what its existence implies about Anthropic’s internal models. Anthropic almost certainly has substantially stronger internal models helping generate training environments. Think of
In a post on X, analyst ChrisGPT argues that while Opus 5.5 has received unanimous praise, its existence implies Anthropic has substantially stronger internal models acting as teachers to generate training environments, with Opus 5.5 as the cheaper deployable student. The post reports that Opus 5.5 scores 55.8% on CoBench 2.1, while Anthropic estimates roughly 85% would be required to fully substitute for its research staff. The analyst notes Opus 5.5 is now only 30 percentage points away from Anthropic's benchmark threshold for fully substituting its research staff, suggesting rapid progress toward that goal.
Claude Opus 5.5 tops Coding Agent Index with score of 66, Anthropic cuts pricing
Claude Opus 5.5 has achieved the top score of 66 on the Artificial Analysis Coding Agent Index, surpassing Opus 5 (60) and Claude Fable 5.1 (62). The model improved across all three evaluations: Terminal-Bench 4.0 rose to 63.1% from 54.5%, DeepSWE v1.1 to 68.4% from 62.5%, and SWE-Atlas-QnA to 66.4% from 62.1%. Anthropic reduced Opus pricing to $4/$20 per million input/output tokens, down 20% from Opus 5, and cache reads to $0.20 per million, down 60%. Despite these cuts, Opus 5.5's Cost per Task is $13.04, up 21% from Opus 5's $10.79, because it uses substantially more tokens—about 15.6 million per task versus 11.4 million. The model extends the Coding Agent Index vs Cost per Task Pareto frontier, as no lower-cost model matches its score.
Read sourceClaude Opus 5.5 becomes first AI model to pass user's code-meets-poetry challenge
In a post on X, user alliekmiller reports that Anthropic's Claude Opus 5.5 has become the first AI model to pass their 'code-meets-poetry' challenge, marking a shift from their previous preference for OpenAI models in ideation, creation, and exploration tasks. The user describes testing Opus 5.5 for slide design, where it generated six PNG versions using OpenAI's image 2.5, labeled them, saved them, opened Finder, and messaged its preference. When the user indicated two preferences, the model autonomously created editable PowerPoint slides with text boxes and shapes. The user notes that earlier tool call failures in Opus 5.5 appear to have been fixed, and characterizes the model as a 'tool-wielding machine' that feels far more creative than previous Claude models.
Show 17 older updatesHide older updates
Claude Opus 5.5 tops writing benchmark with record 307-point Elo gap over second-place model
An X post from the account 'Whats_AI' reports that Claude Opus 5.5 achieved a score of 2631 Elo on a writing benchmark, surpassing the second-place model Fable at 2324 Elo by a 307-point gap. The post describes this as the largest single jump recorded since the benchmark began in June 2026, and notes that Opus 5.5 is the first model to clear 91 out of 100 on the rubric. However, the post also notes that at maximum effort, the model takes 17 minutes and costs $3.43 to write one script, making it the slowest configuration on the board by far and among the top five most expensive. The author states the performance 'almost broke our benchmark' and that they initially thought it was a bug.
OpenAI and Anthropic Launch New AI Models on Same Day, Competing on Cost Efficiency
On September 22, 2025, OpenAI and Anthropic released new AI models on the same day, both emphasizing cost efficiency. OpenAI launched GPT-6 Sol and GPT-6 Luna, with API prices significantly reduced from their predecessors: Sol's input/output prices are $2/$10 per million tokens, half of GPT-5.6 Sol's $4/$20; Luna's prices are $0.10/$0.50, with output down 58% from $1.20. Anthropic released Claude Opus 5.5, priced at $4/$20 per million tokens, 20% lower than Opus 5, with cache reads down 60%. OpenAI claims Sol achieves 33.2% on AutomationBench at $0.27 per task, while Claude Opus 5 scores 26.9% at 11.1x the cost. On DeepSWE v1.1, Sol scores 68.8% near Claude Fable 5's 69.9% at 80% lower cost. Anthropic says Opus 5.5 matches Claude Fable 5.1 on most tasks, scoring 66.4% on Terminal-Bench 4.0 versus Fable 5.1's 55.8%. Third-party testing by Artificial Analysis ranks Opus 5.5 first in reasoning intensity at 58, with GPT-6 Sol at 48 and Luna at 37. The article also notes Meta's AI assistant Muse topped the US App Store, and Xiaomi released MiMo-V2.6 models in China at prices 1/20 to 1/60 of overseas models.
Read sourceOpenAI launches GPT-6 Sol and Luna, Anthropic releases Claude Opus 5.5
The September 23 daily AI brief reports multiple major product launches and updates. OpenAI rolled out GPT-6 Sol and Luna in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu tiers, with API prices 50% below GPT-5.6 promo at Sol $2/$10 and Luna $0.10/$0.50 per 1M tokens. Free and Go users can try Luna in the desktop app. Anthropic released Claude Opus 5.5, the first of the 5.5 family, achieving Fable 5.1-level performance on most tasks while being ~40% cheaper and ~30% faster than Opus 5. Platform pricing is $4/$20 per 1M tokens with cache reads dropping to $0.20. SpaceXAI's Grok in Tesla now runs Connectors for inbox, calendar, files, and chats hands-free, enabling coffee orders, reservations, and appointments. Cursor integrated Opus 5.5, which tops CursorBench at 57.8% Max at 40% less per task than Opus 5. Perplexity made Sol the default Light model and Opus 5.5 the Standard effort. DigitalOcean launched Managed Agents public preview with isolated microVMs and 16k+ tools. Alibaba's Qwen-Image-2.1 ranked #1 open-source in Arena Image Edit and Text-to-Image. Xiaomi shipped MiMo-V2.6 Pro and Flash, with Pro being MIT open-weight. China reportedly opened a data-security probe into DeepSeek and Moonshot after Anthropic alleged Claude traffic routing.
Read sourceOpenAI Releases GPT-6 Sol and Luna Models With 50% Lower API Prices
On September 22, OpenAI released two new AI models, GPT-6 Sol and GPT-6 Luna, with API prices 50% lower than the promotional price of GPT-5.6. GPT-6 Sol charges $2 per million input tokens and $10 per million output tokens, while GPT-6 Luna charges $0.1 per million input tokens and $0.5 per million output tokens. OpenAI stated that GPT-6 Astra remains its best-performing model overall, which was released 19 days ago and described as 'the world's most intelligent and best-aligned model.' On the same day, Anthropic released Claude Opus 5.5, claiming it matches Claude Fable 5.1's performance on most tasks while reducing typical task operating costs by 40% compared to Opus 5. Anthropic stated that Opus 5.5 outperforms OpenAI's GPT-5.6 Sol in software development benchmarks, with operating costs approximately one-third of the latter. Opus 5.5 is priced at $4 per million input tokens and $20 per million output tokens.
Read sourceAnthropic Launches Claude Opus 5.5, Cutting Task Costs by 40% and Boosting Speed
On September 23, Anthropic released Claude Opus 5.5, the first model in its Claude 5.5 series. The company claims the new model matches Claude Fable 5.1 on most tasks while reducing typical task costs by 40% and increasing output speed by over 30% compared to the previous Opus 5. In nine benchmark tests, Opus 5.5 outperformed Fable 5.1, Opus 5, GPT-6 Astra, and GPT-5.6 Sol in seven categories, including three programming tests, but lagged behind GPT-6 Astra in business process and scientific agent tests. On the Artificial Analysis Intelligence leaderboard, Opus 5.5 scored 58 points, ranking first. Pricing is set at $4 per million input tokens and $20 per million output tokens, a 20% reduction from Opus 5. Anthropic highlighted real-world tasks: Opus 5.5 rewrote HAProxy from C to Rust in 9.5 hours at 51% lower cost than Fable 5.1, and passed 16 of 18 financial report generation tests where Fable 5.1 and Opus 5 failed all. The model is available on Claude, AWS, Google Cloud, and Azure. However, most cybersecurity tasks are redirected to Opus 4.8 due to safety restrictions. Claude Sonnet 5.5 and Haiku 5.5 are planned for release in the coming weeks. Shortly after this launch, OpenAI released GPT-6 Sol and GPT-6 Luna.
Read sourceAnthropic Launches Opus 5.5 AI Model, Claims Performance Edge Over Fable at 40% Lower Cost
Anthropic has released Opus 5.5, the first model in its new Claude 5.5 series, with Sonnet 5.5 and Haiku 5.5 expected in weeks. The model is positioned as a direct competitor to OpenAI's GPT-6 Astra, offering API input pricing at $4 per million tokens (40% of Astra's $10) and output at $20 (vs. Astra's $50). In Anthropic's benchmarks, Opus 5.5 outperformed Astra in 4 of 6 comparable tests, including Terminal-Bench 4.0 (66.4% vs. 57.9%) and FrontierCode (54.4% vs. 53.3%), while Astra led in AutomationBench (41.4% vs. 40.0%) and Terminal-Bench-Science (64.6% vs. 58.7%). Third-party testing by Artificial Analysis gave Opus 5.5 a max-effort score of 58 vs. Astra and Fable 5.1 at 53, but noted Opus 5.5 consumes significantly more tokens per task (11.9K vs. 2.7K), making max-effort tasks more expensive ($5.98 vs. $3.26). The model features mandatory adaptive thinking, removed forced tool calling, and improved writing style. Anthropic also implemented enhanced safety measures, including a new test for boundary-crossing behavior. The article notes that Fable 5.1, released only three weeks prior, now faces an awkward position as Opus 5.5 matches or exceeds its capabilities at lower cost.
Read sourceClaude Opus 5.5 costs $5.98 per task, similar to Opus 5 at $5.86 after price cuts
An analysis of Anthropic's pricing reveals that Claude Opus 5.5 (max) costs $5.98 per Intelligence Index task, closely matching the $5.86 cost of Opus 5 (max). However, this price parity is achieved through a combination of factors. Opus 5.5's increased token usage would normally drive an approximately 80% increase in cost per task to $10.51. Anthropic applied a 20% cut to base prices, reducing the cost to $8.41, and further cut cache reads to $0.20, bringing the final cost to $5.98, an additional 29% reduction. The post attributes these price reductions to Anthropic's pricing strategy, which bundles a significant token usage increase with the lower prices.
Read sourceAnthropic Launches Cost-Efficient Opus 5.5 AI Model Ahead of Potential IPO
Anthropic PBC has released a new AI model, Claude Opus 5.5, designed to handle a broad range of enterprise tasks at a significantly lower cost than its predecessor, Opus 5. The company claims the model matches the performance of its more powerful Fable 5.1 on most work tasks while being faster and producing clearer text. This launch comes as Anthropic prepares for a highly anticipated Wall Street debut, with an IPO expected as early as this fall, ahead of rival OpenAI's planned 2026 IPO. The company is also planning to release new versions of Claude Sonnet and Haiku in the coming weeks to expand the 5.5 series. CEO Dario Amodei, who has called for the industry to slow the pace of AI development, stated that the company has implemented new safety measures, including third-party evaluations by METR and Frontier Design, and deployed safety guardrails similar to those on the more powerful Fable model to limit high-risk cybersecurity and biological applications.
Anthropic Opus 5.5 beats Fable 5.1 on all benchmarks; OpenAI GPT-6 Luna scores 66.6% on DeepSWE 1.1
A news intelligence analyst recaps the day's AI releases from Anthropic and OpenAI, concluding there is no clear winner. Anthropic addressed user feedback on Opus communication and introduced a subscriber reset. Opus 5.5 outperforms Fable 5.1 on every benchmark in Anthropic's comparison table, delivering Fable 5.1-level performance at roughly 40% lower cost than Opus 5. Sonnet 5.5 and Haiku 5.5 are expected in coming weeks. OpenAI's GPT-6 Luna achieved 66.6% on DeepSWE 1.1 at max effort, comparable to Fable 5 at medium effort but at 96% lower cost per task. Luna's API pricing is $0.10 per million input tokens and $0.50 per million output tokens. GPT-6 Sol approaches Astra's factual reliability on internal evaluations and outperforms low-effort Astra on AutomationBench at xhigh effort. The analyst also notes unverified rumors about an OpenAI counterpart to Grok Bot.
Anthropic and OpenAI Launch New AI Models, Intensifying Price War
On September 22, Anthropic released the Claude Opus 5.5 model, claiming performance comparable to its flagship Fable 5.1 at 40% lower cost than Opus 5, with pricing at $4 per million input tokens and $20 per million output tokens, a 20% reduction. The model scored higher than OpenAI's GPT-5.6 Sol on the CursorBench programming benchmark at one-third the cost. OpenAI simultaneously launched GPT-6 Sol and GPT-6 Luna. GPT-6 Sol is priced at $2 per million input tokens and $10 per million output tokens, 50% cheaper than GPT-5.6 Sol's promotional price, with half the error rate. GPT-6 Luna costs $0.10 per million input tokens and $0.50 per million output tokens, matching GPT-5.6 Sol's performance at high reasoning intensity for about 1% of the cost. OpenAI claims GPT-6 Sol outperforms Anthropic's Claude Opus 5 by 6.3% on the AutomationBench test at 9% of the cost. Both companies highlighted safety improvements: Anthropic said Opus 5.5's probability of attempting to breach isolation boundaries is 85% lower than prior models, while OpenAI reported reduced rates of bypassing restrictions. The releases come amid investor speculation that Anthropic's new model could precede its IPO filing, as the market watches for growth amid price-sensitive customers.
Read sourceAnthropic Launches Cheaper Opus 5.5 AI Model Ahead of Anticipated IPO
On September 23, Anthropic announced the launch of a new AI model, Claude Opus 5.5, designed to handle a variety of business tasks at a lower cost. This move is part of the company's strategy to maintain a competitive edge ahead of its highly anticipated Wall Street IPO. According to Anthropic, the Opus 5.5 model delivers performance comparable to its more powerful Fable 5.1 model in most work scenarios, while being significantly cheaper to use than the previous Opus 5 version. The company also stated that the new model offers faster output speeds and produces clearer content. Anthropic plans to expand its 5.5 series in the coming weeks with new versions of Claude Sonnet and Haiku.
Read sourceAnthropic Launches Cheaper Claude Opus 5.5 AI Model Ahead of Anticipated IPO
Anthropic has released a new AI model, Claude Opus 5.5, on Tuesday, designed to handle a variety of business tasks at a lower cost. The company claims the model's performance is comparable to its more powerful Fable 5.1 in most work scenarios, while being significantly cheaper to use than the previous Opus 5 version. Additionally, Opus 5.5 offers faster output speeds and generates clearer content. This launch is part of Anthropic's strategy to maintain a competitive edge ahead of its highly anticipated Wall Street IPO. The company plans to expand its 5.5 series in the coming weeks with new versions of Claude Sonnet and Haiku.
Read sourceAnthropic launches Claude Opus 5.5 AI model, claiming lower cost for enterprise tasks
Anthropic released its new AI model, Claude Opus 5.5, on Tuesday, designed to handle a wide range of enterprise tasks at a lower cost. According to Anthropic, the model performs similarly to the more capable Fable 5.1 on most work tasks, while being significantly cheaper than the previous Opus 5. The new model also generates results faster and produces clearer text. Anthropic plans to expand the 5.5 series in the coming weeks with new versions of Claude Sonnet and Haiku. Opus 5.5 is the first new model released since Anthropic CEO Dario Amodei called for the AI industry to slow down the pace of technical development. Amodei previously stated that the company would adopt new safety measures, including introducing third-party evaluators.
Read sourceClaude Opus 5.5 tops Artificial Analysis Intelligence Index, beating Fable 5.1 and GPT-6 Astra
In a post on X, user Mr_Salio announced that Claude Opus 5.5 has achieved the number one position on the Artificial Analysis Intelligence Index. The post claims that Opus 5.5 outperformed both Claude Fable 5.1 and GPT-6 Astra in the ranking, using informal language to describe the competitive result. Additionally, the post asserts that Opus 5.5 is significantly cheaper than both competing models. The announcement includes a link to the Artificial Analysis website for further details. The post attributes the ranking to the Artificial Analysis Intelligence Index, an external benchmark, and presents the information as a factual development in the AI model landscape. No further context or methodology is provided in the post itself.
Anthropic's Opus 5.5 cuts AI coding bills by 40%, outperforms GPT-6 Astra at 20% cost
Aakash Gupta reports that Anthropic's new Opus 5.5 model will significantly reduce AI coding costs, far exceeding the commonly cited 20% price drop on regular tokens. The key savings come from a 60% reduction in the cost of re-reading files, which constitutes most of an AI agent's expenses. Additionally, the model requires fewer steps to complete tasks, with one early tester fixing a 200,000-line codebase in under 3 hours compared to over 20 hours for the old Opus. Anthropic states the typical bill drops about 40% when combining these factors. In a test measuring whether AI code would be approved by a real team, Opus 5.5 on its normal setting beat the best score ever posted by OpenAI's GPT-6 Astra, at roughly a fifth of the cost per task. Gupta notes this is the first time in the AI race that the top-performing coding model is also the cheaper one to run, posing a challenge for OpenAI to respond with a price cut.
Claude Opus 5.5 launches claiming Fable 5.1-level performance with 40% cost cut
Anthropic has released Claude Opus 5.5, a new AI model that claims to deliver performance comparable to Fable 5.1 while reducing typical workload costs by 40%. The pricing for input and output tokens has been set at $4 and $20 per 1 million tokens respectively, representing significant reductions compared to Opus 5. Cache read costs have dropped 60% to $0.20. Additionally, the model's output arrives more than 30% faster, with a Fast mode option that reaches up to 2.5x speed at double the token prices. These updates position Claude Opus 5.5 as a more cost-effective and efficient alternative in the AI model market.
Read sourceAnthropic's Opus 5.5 scores 58 on Artificial Analysis Intelligence Index, topping Fable 5.1 and GPT-6 Astra
A post on X reports that Anthropic's AI model, Opus 5.5, has achieved a score of 58 on the Artificial Analysis Intelligence Index. This score surpasses the next highest scores in the chart, which are 53 for both Fable 5.1 and GPT-6 Astra. The post expresses surprise at the performance, questioning what Anthropic used to train the model. The source is a user named DanDr1s, and the post includes a link to further information. The claim is presented as a new development in AI model benchmarking.
Read source