← The Polylog AI Intelligence Brief
Morning Edition · Monday, July 27, 2026Published at 1:32 AM EDT · New York
Anthropic Ships Claude Opus 5, Priced at Half Its Largest Model
Opus 5 lists at $5 per million input tokens and $25 per million output tokens. On Anthropic's own numbers, it roughly doubles its predecessor's score on an agentic-coding benchmark.
Anthropic launched Claude Opus 5 on July 24, positioning it as the model for long-running agents, with improvements in coding and professional work. It costs $5 per million input tokens and $25 per million output tokens, which third-party writeups describe as roughly half the input price of the larger Claude Fable 5 while approaching that model's intelligence.
The reported gains are concentrated in agentic tasks. Coverage citing Anthropic's figures puts Opus 5 at 79.2% on SWE-bench Pro and about 43.3% on Frontier-Bench, more than double Opus 4.8, with an ARC-AGI-3 score roughly three times the next-best model. As with any launch, these are vendor-run numbers. The claim engineers can test directly is cost per completed task, where one benchmark put Opus 5 within half a percent of Fable 5 at half the price.
The release comes two days before Moonshot's Kimi K3 weights, and the positioning is telling. Anthropic is competing on price for each unit of capability at the Opus tier rather than pushing a new performance ceiling. That indicates the competition in coding models now turns on unit economics as much as on peak scores.
What this means
Pricing a near-frontier Opus tier at half the flagship's cost is a direct move in the competition over inference costs. It pressures OpenAI's and Google's mid-tier coding models on dollars per resolved ticket, the measure enterprise buyers actually budget against. The exposed parties are competing closed labs whose comparable tiers cost more. The beneficiaries are agent platforms and Claude Code users whose long-running processes consume large numbers of output tokens. The open question is whether the doubled agentic scores survive independent SWE-bench Verified runs or shrink under standardized testing.
What to watch
- Independent agentic-coding leaderboards that report Opus 5 cost per task, which will confirm or undercut Anthropic's claim of price parity with Fable 5.
- Whether OpenAI and Google respond with price cuts or new mid-tier coding models within weeks, which would signal that profit margin, not capability, now drives the competition.
Part of a tracked trend
Frontier Labs Race on AI Coding Capability
Coding is becoming a primary competitive battleground among frontier labs, with incumbents standing up permanent coding teams and investing in new training stages (e.g. midtraining) to match leaders like Anthropic; expect recurring reorganizations, benchmarks, and model releases aimed specifically at code.
Townsquare
Start a discussion in Townsquare.
More from this edition
- Moonshot AI Releases Kimi K3 Weights, a 2.8-Trillion-Parameter Open Model
- Nvidia Uses Its Own Vera CPU to Speed Up Chip Design by 1.5 Times
- OpenAI Puts ChatGPT Voice in the Desktop App to Drive Codex Agents
- SenseTime Open-Sources SenseNova-Vision, a Unified Perception Model
- Meta's Brain2Qwerty v2 Decodes Typed Sentences at 61% Word Accuracy
- FlowEvo Lets LLM Agents Co-Evolve Workflows and Reusable Skills
- CARE Proposes Pre-Execution Verification for Shell-Executing LLM Agents
- Paper Uses Reinforcement Learning to Optimize Stylistic Jailbreaks of Vision Models
Comments
0No comments yet.