Anthropic Launches Claude Opus 5 with Autonomous Reasoning at Stable Cost

Anthropic released Claude Opus 5 on July 27, 2026, offering frontier intelligence at half the cost of Claude Fable 5 while maintaining the same token rates as Claude Opus 4.8—$5 per million input tokens and $25 per million output tokens. The model autonomously decides when to engage extended reasoning, features a 1M-token context window, and is available immediately on the Claude API, AWS, Google Cloud, and Microsoft Foundry.

July 27, 2026, Inside AI — Anthropic released Claude Opus 5 today, positioning it as a significant leap in reasoning and cost efficiency. The model arrives with a notable pricing strategy: it delivers frontier intelligence at half the cost of Claude Fable 5 , while maintaining the same token rates as its predecessor, Claude Opus 4.8 — $5 per million input tokens and $25 per million output tokens. This pricing move challenges the typical premium-tier model, offering a meaningfully stronger model without a price hike. For developers and businesses, the immediate draw is enhanced capability at a stable cost, a rarity in the competitive AI landscape. The core advancement is a shift in reasoning mechanics. Unlike Opus 4.8, where extended reasoning required manual activation, Opus 5 autonomously decides when and how deeply to think on each turn. This should improve out-of-the-box answers, but it also forces developers to adjust output-token limits, as thinking and response text now share the same budget. Opus 5 converts extra effort into better results more reliably across a ladder from low to a new "max" tier for the deepest reasoning. However, there’s a deliberate design catch: disabling thinking works only at high effort or below. Requesting no thinking at top tiers returns an error, enforcing a minimum reasoning threshold. Anthropic highlights stronger agentic coding—completing multi-file features without stubs, sharper code review with fewer false positives, and improved vision for charts, documents, and UI work. The model also features a 1M-token context window as standard, alongside better spreadsheet and slide generation. These capabilities build on research into long-context models, as explored in works like efficient long-context understanding https://arxiv.org/abs/2402.19450 . Behavioral changes are subtle but critical for upgrades. Opus 5 writes longer responses by default, narrates its progress more during agentic sessions, and verifies its own work without prompting. Anthropic even advises removing old "add a verification step" instructions, as they now cause over-checking. This self-verification aligns with trends in self-improving AI systems https://arxiv.org/abs/2303.17651 . The launch intensifies competition with models like OpenAI’s o-series and Google’s Gemini, which also emphasize reasoning and agentic tasks. Anthropic’s focus on autonomous reasoning depth and cost stability may pressure rivals to rethink pricing structures. However, the forced thinking at high tiers could limit flexibility for users who need rapid, shallow responses in cost-sensitive applications. Opus 5 is available immediately on the Claude API , AWS , Google Cloud , and Microsoft Foundry . Anthropic’s decision to keep pricing steady while boosting performance signals a strategic push to capture market share in enterprise AI, where long-horizon reliability and coding prowess are paramount.