Anthropic's Claude Opus 4.8 is here with 3X cheaper fast mode and near-Mythos level alignment
Anthropic launched Claude Opus 4.8 with unchanged standard pricing, a much cheaper fast mode, and new parallel subagent workflows for large code tasks.
Intelligence analysis by GPT-5.4 Mini
Claude Opus 4.8 is a measured upgrade that cuts fast-mode costs, adds large-scale agent workflows in Claude Code, and narrows the gap to Anthropic's more restricted Mythos model on alignment.
Anthropic made a new version of its smart helper called Claude Opus 4.8. It can answer faster in a cheaper mode, and it can also split big computer jobs into lots of smaller helpers working at the same time.
Think of it like a school project where one student reads, another writes, and another checks the math instead of one child doing everything alone. That can help with huge coding jobs.
Anthropic also says this version makes fewer mistakes than before and acts more carefully. That matters because companies want a helper that is fast, cheaper to run, and less likely to cause trouble.
Analysis
Pricing and access
Anthropic says Claude Opus 4.8 is available immediately across claude.ai, Claude Code, the API, and Cowork, with standard pricing unchanged at $5 per million input tokens and $25 per million output tokens. The larger commercial shift is fast mode, which Anthropic cut to $10 per million input tokens and $50 per million output tokens, down from $30 and $150 on Opus 4.7. The company says fast mode runs about 2.5x faster than normal mode and is intended for higher-throughput workloads. In Claude Code, users can trigger it with /fast, while API access is waitlisted.
Capability gains
Anthropic describes Opus 4.8 as a modest upgrade rather than a leap, but the benchmark gains are real: SWE-bench Verified rises to 88.6%, SWE-bench Pro to 69.2%, and Terminal-Bench 2.1 to 74.6%. The company also says the model beats OpenAI's GPT-5.5 regular on at least 12 benchmarks, especially in knowledge work, coding issue resolution, agentic tool use, and long-context tasks, while GPT-5.5 remains stronger on terminal workflows.
Bigger workflow shift
The release also introduces a research preview of dynamic workflows in Claude Code. Anthropic says the system can plan work, spawn hundreds of parallel subagents, and verify results before returning an answer, aimed at tasks too large for one context window. The example given is a codebase-wide migration across hundreds of thousands of lines. Separate additions include an effort selector in claude.ai and Claude Cowork, plus API support for updating system instructions mid-task without breaking the prompt cache.
Alignment and enterprise signal
Anthropic is also framing Opus 4.8 as more reliable and better aligned. The company says it is around four times less likely than its predecessor to let code flaws go unremarked, and that misaligned behavior is now substantially lower than Opus 4.7 and close to Claude Mythos Preview. Several enterprise customers, including Databricks, Hebbia, and Cognition, say the model improves reasoning, citation quality, token efficiency, and tool use. Anthropic says Mythos-class models should reach all customers in the coming weeks once more cyber safeguards are in place.
Key points
- Claude Opus 4.8 ships at the same standard price as Opus 4.7.
- Fast mode is much cheaper than before, with Anthropic saying it is now 3x less expensive.
- Claude Code now has dynamic workflows that can spawn hundreds of parallel subagents for large code tasks.
- Anthropic says Opus 4.8 improves benchmark scores and alignment versus Opus 4.7.
- Enterprise customers cited gains in reasoning, citation accuracy, and token efficiency.
