Claude 4 Opus Opens to API: 1M Context, $15/M Tokens
Anthropic has made Claude 4 Opus generally available via API after a limited preview, offering its highest-capability model with a 1M-token context window and improved instruction following at $15 per million input tokens.
Original sourceAnthropic has moved Claude 4 Opus from limited preview to general availability for API customers, opening its most capable model to production workloads. The model ships with a 1 million token context window, putting it in direct competition with Gemini 1.5 Pro and GPT-4o on long-context tasks. Pricing is set at $15 per million input tokens and $75 per million output tokens, positioning it at the premium end of the frontier model market.
The general availability release follows a restricted period during which select partners and enterprise customers could access the model. Anthropic emphasizes improvements in instruction following and complex reasoning relative to earlier Claude versions, making it a target for agentic workflows, large document analysis, and multi-step coding tasks where context fidelity matters.
The 1M-token context window is the headline technical differentiator here. At that scale, entire codebases, legal documents, or research corpora can be loaded in a single prompt. Whether the model maintains coherent attention across the full context — a known challenge for all long-context models — will determine how much of that headline translates to real-world utility. Anthropic has not published a needle-in-a-haystack benchmark at the 1M boundary in this release announcement.
At $15 per million input tokens, Opus sits above Claude Sonnet and Haiku in Anthropic's tiered lineup. For teams already on the Anthropic platform, the path to upgrading is a model string swap. For new customers, this is an entry point into Anthropic's ecosystem at a price point that rewards high-value, low-volume use cases over high-throughput commodity inference.
Panel Takes
The Builder
Developer Perspective
“The primitive here is a long-context frontier model with a clean REST API — and for Anthropic that means the same SDK, same auth, same request shape, just a new model string. That's the right DX bet: zero migration cost for existing customers. What I want before trusting this in production is a real needle-in-a-haystack result at 800K+ tokens, not a marketing claim — Anthropic didn't ship one with this announcement and that's a gap worth noting before you redesign your chunking strategy around a 1M window.”
The Skeptic
Reality Check
“The 1M context window claim is doing a lot of work here without a single published benchmark to back it up at that scale — every major lab has shipped "1M context" and quietly watched coherence fall off a cliff past 200K in real retrieval tasks. At $15 per million input tokens, a 1M-token prompt costs $15 per call before you generate a single output token, which means this pricing structure only survives contact with users who have very specific, very high-value use cases. What kills this in 12 months isn't a competitor — it's Anthropic itself shipping Sonnet-class performance at Haiku-class prices and leaving Opus as a niche tax on enterprise procurement.”
The Futurist
Big Picture
“The thesis Anthropic is betting on: by 2027, the bottleneck in agentic software isn't model intelligence — it's context capacity, and whoever owns the 1M-token standard owns the infrastructure layer for long-running autonomous workflows. That's a plausible bet, but it only pays off if context utilization actually scales — meaning models don't just accept a million tokens but reliably act on information buried at position 600K. The second-order effect if this works is that traditional RAG pipelines become legacy overnight, redistributing power from vector database companies and retrieval middleware back to the raw model API.”
The Founder
Business & Market
“The buyer here is clear — enterprise teams with legal, compliance, or codebase-scale document workflows where GPT-4o's context limits are already a friction point, and the check comes from an AI infrastructure or engineering budget. The moat concern is real: Anthropic's defensibility isn't the 1M window, which Google and OpenAI have matched or will match, it's Constitutional AI and the enterprise trust narrative around safety that justifies procurement approval in regulated industries. What I'd watch is whether the output token price of $75 per million creates a ceiling on agentic use cases — any workflow that generates substantial output at scale will hit economics that push buyers back toward Sonnet.”