ANTHROPIC · MODEL RELEASE TRACKER
Claude Opus 5
Claude Opus 5 is an Anthropic model announced as available on 2026-09-16. Anthropic describes it as a thoughtful, proactive model that approaches the frontier intelligence of Claude Fable 5 at roughly half the price, with especially strong results on coding and knowledge-work benchmarks and meaningful gains over Opus 4.8. It is the new default on Claude Max and the strongest model on Claude Pro, and Anthropic notes it remains behind Mythos 5 on cybersecurity tasks.CURRENT SNAPSHOT5/5 DIMENSIONS WITH DATA
The dimensions that change the decision.
RELEASE TIMELINE
Published, source-backed release events only.
Cloudflare launches Auto Router public beta in AI Gateway to rate requests by stakes and other metrics
Cloudflare published Auto Router in public beta via AI Gateway: a two-stage router that builds a candidate model pool then runs a multi-head classifier on Workers AI to assign probabilities across 14 task categories and score requests 1–5 on complexity, ambiguity, stakes and context dependence, using only the newest turns (up to 8 KiB). The router combines those scores with benchmark results (quoted formula: "utility = expected quality - adaptive cost penalty"), exposes routing headers for auditability, supports per-request allowed-models headers, is free in beta, and currently lacks WebSocket support.
Nvidia open-sources Flora, a Qwen3.5-based robot video labeling engine
Nvidia's GEAR lab open-sourced Flora, a data-engine composed of three fine-tuned Qwen3.5 models (9B for segmentation, 27B for descriptions, and a 27B 'Flora-Rewarder' for scoring) that labels robot videos for action boundaries, step-level rewards, and task success. Nvidia says Flora is already used in training its GR00T models at scale on GB300 GPUs and estimates labeling costs of about $0.72/hour versus ~$88/hour with GPT-6 Astra, though the cost figures are projections and benchmark comparisons include methodological caveats.
Deepseek releases Deepseek‑V4.1‑Flash‑1017, open-sources weights and opens search APIs
Chinese company Deepseek released Deepseek‑V4.1‑Flash‑1017 and published the model weights on Hugging Face and ModelScope. The company says the model—trained with reinforcement learning using self-feedback—runs faster, produces shorter outputs, beats larger rivals in its internal coding and agent benchmarks, and is paired with new Search & Fetch APIs for developers; independent verification of the benchmark claims is not yet available.
Anthropic releases Claude Opus 5 — lower-cost model claiming near‑frontier performance
Anthropic announced Claude Opus 5, a new model positioned as a cost‑efficient successor to Opus 4.8 and the new default on Claude Max (and the strongest on Claude Pro). Anthropic says Opus 5 matches or exceeds prior models on many coding, knowledge‑work and scientific benchmarks (Frontier‑Bench, GDPval‑AA, CursorBench, ARC‑AGI, Zapier AutomationBench, OSWorld) at lower cost per task while remaining behind Mythos 5 on security and biology frontier tasks; the company also reports improved alignment and safety in pre‑deployment audits and links a System Card for more details.
VERIFIABLE FACTS
Every value stays attached to a source and date.
2026-07-24
1,000,000 tokens
128K tokens
Text, images
Text
Active on Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS. Active (legacy)
$5 USD / 1M input tokens; Claude API standard, global routing
$25 USD / 1M output tokens; Claude API standard, global routing
$0.5 USD / 1M cached input tokens; Claude API standard, global routing
Announced available on 2026-09-16 (Anthropic announcement: 'Claude Opus 5 is available today').
Described as coming close to the frontier intelligence of Claude Fable 5 while costing about half as much.
Reported estimate of about USD 42 per hour for labeling in the same comparison where Flora was estimated at about USD 0.72/hour and GPT-6 Astra at about USD 88.50/hour.
Described as 'the new state-of-the-art' on coding and knowledge-work evaluations such as Frontier-Bench and GDPval-AA.
Anthropic states Opus 5 remains behind Mythos 5 on cybersecurity tasks.
Reported to provide greatly improved performance for the same cost as its predecessor Opus 4.8.
Set as the new default model on Claude Max and described as the strongest model on Claude Pro.
In NVIDIA’s SWE-Serve evaluation, Claude Opus 5 is listed among top performers with a reported pass@1 of 75% on the evaluated inference-engineering tasks.
Anthropic reports Opus 5 as the new state-of-the-art on coding and knowledge-work evaluations (e.g., surpasses all models on Frontier-Bench v0.1, more than doubles Opus 4.8’s performance at lower cost per task; on CursorBench 3.2 it reaches within 0.5% of Fable 5’s peak at about half the cost per task; on ARC-AGI 3 its score is ~3× the next-best model; on Zapier AutomationBench its pass rate is ~1.5× the next-best for the same cost; on OSWorld 2.0 it outperforms every other model at a given cost).
Anthropic states Opus 5 remains behind Mythos 5 on cybersecurity tasks.
In a controlled research task, when constraints were stated directly, Claude Opus 5 recovered the best-supported assignment in 96% of cases (reported by the arXiv study).
The arXiv study reports that in scientific prose Opus 5 often prefers the biologically expected assignment; when that biological preference is removed, recovery of the better-supported assignment increases from 27% to 79%, and reaches 92% when the Biology-favored record is accompanied by a formalization request and explicit paired-design cue.
Reported as 46.67% Pass@3 on the UI-TAU benchmark (as stated in the cited paper).
Reported to have "scored much worse" than Nvidia's Flora (and worse than GPT-6 Astra) in Nvidia's 50-video labeling test.
MARQ achieved a median F1 of 0.80 for Claude Opus 5 on an expert-labeled NASA requirement set (100 executions), compared with an off-the-shelf baseline F1 of 0.55.
In the MARQ evaluation, Claude Opus 5 improved defect detection and false-alarm control compared with other evaluated models.
Criterion-level analysis reported in MARQ showed that gains were concentrated in Correctness and Conformance, indicating model-dependent responses across several criteria.
Reached all required outcomes in 81.6% of its runs on the BABA-Bench evaluation.
Only 52.5% of its runs passed all checks that verify required procedures (i.e., procedural compliance) in the BABA-Bench evaluation.
Claude Opus 5 was one of nine AI agents evaluated in the Carnegie Mellon–led 'Into the Unknown' benchmark, which asked agents to carry out follow-up research studies on open questions from US national labs.
Reported to take the next spots after GPT-6 Astra, alongside Qwen3.8-Max, on the Co-Design Bench coding-agent benchmark.
Mentioned as a US model whose coding-benchmark performance level was matched by LongCat-2.5 a few months later.
WHAT CHANGED