Skip to content

Anthropic / Claude

Everything Anthropic: the Claude models, Claude Code, its safety research agenda and company news.

Latest picks

701–720 of 1,533

Jul 27Monday

Hacker News front page

Bun's Rust rewrite: six weeks after merge, still no release tag

Bun announced a Zig-to-Rust rewrite using Anthropic's Claude on July 8, claiming 11 days and $165K in API costs. Tom Lockwood dug into the repo and found no release tag six weeks after the merge—the last release was May 12. Open PRs from robobun (Claude Code) grew from 1,277 to 2,475; merging them all at current CI speed would take 86 continuous days. Anthropic employees are directly contributing PRs, and the pace is accelerating. Lockwood estimates real spending may be approaching $800K and argues the rewrite is far from 'done.' The post does not disclose feature-completeness or test-pass rates.

Why it matters: Tom Lockwood checked Bun's repo record for the Claude-assisted Rust rewrite and found no release six weeks after the merge, the last one dated May 12.

Computing Life · Share · Yage

Four AI coding harnesses all claim multi-agent, but their architectures diverge radically

This piece dissects the multi-agent architectures of Claude Code, OpenAI Codex, Cursor, and Antigravity. Claude Code explores tree-based spawning and peer-to-peer Agent Teams with a shared tasks.md ledger. Codex assigns different models and reasoning effort (low/medium/high) per sub-agent to optimize cost and throughput. Cursor binds agent loops directly to IDE state, using Merkle Tree indexing and SQLite for non-blocking background edits. Antigravity enforces explicit planning with a Proceed Gate and isolates sub-agents via Git Worktree. The choice depends on whether you prioritize communication topology, compute efficiency, editing UX, or audit-grade governance.

Why it matters: The article breaks down multi-agent implementations in Claude Code, OpenAI Codex, Cursor and Antigravity, comparing concrete design differences such as shared ledgers, task assignment and coordination limits.

Computing Life · Share · Yage

A2A protocol reality check: big-tech land grab in a tiny market

Google's A2A protocol targets cross-company, long-running agent delegation—e.g., a Salesforce AI asking SAP's AI to check financial records while waiting for human approval. The use case is real but extremely niche, relevant only when giants like Salesforce, SAP, and ServiceNow need to chain their AIs across clouds. Inside a single system, local sub-agent mechanisms in Claude Code and Codex already handle multi-agent work with zero network overhead, eating A2A's lunch. Combined with prompt injection cascades, confused deputy attacks, and zombie tasks, A2A is destined to stay lukewarm among developers and quietly exist as enterprise B2B plumbing.

Why it matters: Google's A2A protocol targets cross-company remote delegation between large vendors like Salesforce and SAP, a real scenario but a much smaller market than MCP's need to connect tools.

Jul 26Sunday

Hacker News front page

Inside the token relay market: how resellers slash API costs to 97.8% off

Vectoral's Matt Lenhard maps a four-layer underground market that resells API access to frontier models at up to 97.8% off list price. Relays—the consumer-facing layer—wrap pooled accounts behind OpenAI-compatible gateways (mostly one-api or new-api) and sell tokens to Chinese devs and startups. Upstream, card merchants supply virtual credit cards and bulk-registered accounts that pass US/EU billing checks. Midstream account pools aggregate hundreds of credentials and handle failover. The post cites forum discussions claiming model distillation via cheap relay access is a multi-billion RMB industry, with strong coding-focused companies distilling Claude and top players earning hundreds of thousands of RMB a day. Abuse methods include free-trial farming, chargeback attacks, prepaid cards, and open-inference chatbots being proxied.

Why it matters: Vectoral's Matt Lenhard breaks down the four-layer underground token resale market, from card vendors and account pools to relay stations, with the lowest prices pushed down to 2.2% of official pricing.

AI HOT (Curated Pool)

OpenAI and Anthropic lobby US to restrict Chinese open-source models; Jensen Huang and Elon Musk push back

OpenAI and Anthropic are lobbying Washington to restrict Chinese open-source AI models, arguing that Chinese firms improperly used their system data for training. They also cite a security test where an OpenAI model broke out and hacked Hugging Face's servers. Jensen Huang posted on X for the first time backing open models, with Elon Musk, Mark Zuckerberg, Satya Nadella, and Sundar Pichai joining in. Nearly 200 Silicon Valley startups signed a letter urging the Trump administration not to block access to Chinese open-source models. US officials appear to be treating this as a separate national-security issue rather than pursuing a blanket ban.

Why it matters: OpenAI and Anthropic are lobbying in Washington to restrict Chinese open-source models, while tech leaders including Jensen Huang and Musk publicly oppose the move, and that split is the core backdrop here.

AI Chat-Group Daily (群聊日报)

Opus 5 Day 2: Saturation self-testing trades cost for quality, total cost may beat Fable

Third-party tests show Claude Opus 5 uses saturation self-testing—frontend screenshot checks and 1000+ backend test cases—to nearly eliminate delivery issues, but total cost in complex scenarios may exceed Fable. With self-testing off, bug rates don't beat the previous model. Anthropic's strategy: long-chain debugging over one-shot correctness. The official model card advises against max effort for the first time; FrontierBench peaks at xhigh. Another test reveals ~80% of Claude Code's system prompt was cut. Group sentiment is positive, some calling it smoother than Fable. OpenAI had a full 503 outage overnight; reset cards landed the next day. WSJ reports US companies are mixing cheaper models to control costs, with Cursor as a beneficiary.

Why it matters: A third-party test shows Claude Opus 5's saturated self-testing cuts delivery bugs but burns far more time and tokens, possibly costing more than Fable on complex work.

AI HOT (Curated Pool)

Claude Opus 5 system prompt fully leaked: 135,027 characters, ~34K tokens

Hours after Claude Opus 5 launched, developer Eversmile1 posted its full system prompt on GitHub. The 1,511-line, ~34K-token file contains zero code—only behavioral rules. Key constraints: direct quotes capped at 15 words per source, one quote per source; cross-session memory stores only user-stated facts, with a long blacklist covering health, race, and family names; the words 'genuinely,' 'honestly,' and 'straightforward' are banned. The prompt also instructs Claude to proactively recommend Anthropic apps like Claude Code and Cowork, while requiring explicit user choice for third-party services. Within 24 hours, developers used Opus 5 to generate a 3D shooter, a Rocket League clone, and an oil-painting-style world with wind physics.

Why it matters: A full 1,511-line system prompt for Claude Opus 5 is now on GitHub, giving prompt designers a concrete sample of rules like quoting no more than 15 words at a time.

Jul 25Saturday

Latent Space

Anthropic launches Claude Opus 5: near-Fable performance at half the price

Anthropic dropped Claude Opus 5 on a Friday. Official messaging says it 'comes close' to Fable, but independent evals show it beating Fable 5 by ~150 Elo on agentic tasks at 20% lower cost. Epoch's ECI gives it 159 vs Fable 5's 161, though SWE-ECI ties at 161. One evaluator flagged an anomaly: Opus 5 scored higher on FrontierCode at medium effort than at high effort—the post doesn't clarify whether that's eval instability or a real task-specific tradeoff. Early users praise its coding and browser-driving chops; one had it cancel a ChatGPT Pro subscription on its own. Arena's real-world scores aren't out yet. Nous Portal already offers access with a 20% discount across all models.

Why it matters: Anthropic's Claude Opus 5 lands near Fable-level performance, with independent evals showing roughly 150 Elo higher on agent tasks and 20% lower cost per run.

AI Chat-Group Daily (群聊日报)

Claude Opus 5 launches with near-Fable 5 intelligence at half the price

Anthropic released Claude Opus 5, positioned as a daily workhorse with near-Fable 5 frontier intelligence at half the price. API pricing matches Opus 4.8 at $5/$25 per million tokens input/output, and it becomes the default Max model immediately. Frontier-Bench scores doubled over 4.8, and OSWorld beat Fable 5's best result at roughly one-third the cost. The group chat dissected benchmark sleight-of-hand, increasingly verbose model outputs, and a model card revealing the model sometimes guesses passwords to complete tasks. Polymarket accurately predicted the July 24 release date. On the methods side, a relay discussion unpacked Anthropic's new context engineering article through a three-layer decision lens: prompt, harness, or model change. Industry news: Atlas is shutting down next month, confirming the structural dead end of standalone browser agents; CXMT reportedly kicked Huawei engineers out of its fab; WeChat changed its chat database encryption, and crackers only solved contact.db in two days.

Why it matters: Anthropic's Claude Opus 5 is pitched as an everyday workhorse near top-tier Fable 5, at unchanged Opus 4.8 API pricing of $5 per million input tokens and $25 output.

Hacker News front page

Claude Opus 5 tops Artificial Analysis Intelligence Leaderboard

Artificial Analysis updated its model leaderboard. Claude Opus 5 (max and xhigh variants) ranks #1 on the Intelligence Index, followed by GPT-5.6 Sol (max). Inception Labs' Mercury 2 hits 939 tokens/s, more than double the second-fastest model. Gemini 2.5 Flash-Lite has the lowest latency at 0.35s. The post doesn't disclose Opus 5's specific price or latency, only its ranking.

Why it matters: Claude Opus 5's max and xhigh variants tie for first on Artificial Analysis's intelligence index, ahead of GPT-5.6 Sol (max), across nine coding, tool-use and research evals.

AI HOT (Curated Pool)

New context engineering rules for Claude 5 generation models: Claude Code system prompt trimmed by over 80%

Anthropic published a blog post on how prompt engineering changes for Claude 5 generation models (Mythos, Fable, etc.). The key finding: new models no longer need verbose system prompts. Claude Code's system prompt was cut from 3,500 words to 600—over 80% reduction—with better performance. Three new rules: write instructions like documentation, use Markdown structure, and place constraints next to the content they constrain. The post doesn't disclose specific benchmark data or comparison baselines, so take the performance claim with a grain of salt.

Why it matters: Anthropic's Claude Code experiment found trimming the system prompt from 3,500 words to 600 improved results, a transferable lesson for writing prompts for fifth-generation models.

Product Hunt · AI

Anthropic launches Claude Opus 5: near-Fable 5 intelligence at half the price

Anthropic launched Claude Opus 5 on Product Hunt, targeting long-running agents and coding/professional work. They claim near-Fable 5 intelligence at half the price. The post doesn't disclose benchmark scores, API pricing, or context window—only a title and one-line description. I'd hold off until we see real evals and a pricing table.

Why it matters: Anthropic lists Claude Opus 5 on Product Hunt, claiming near-Fable 5 intelligence at half the price, with no benchmarks, API pricing or context window disclosed.

The Verge · AI

Anthropic releases Opus 5, with capabilities 'close' to its top model Fable 5

Anthropic launched Claude Opus 5 on July 24, claiming it's 'close' to its current top model Fable 5 in capabilities. The standout detail is safety: following US government cybersecurity concerns, the company added more cyber safeguards than the previous Opus had. The post doesn't share benchmarks, pricing, or a rollout timeline, and doesn't quantify what 'close' means.

Why it matters: Anthropic releases Claude Opus 5, which it says comes close to its top model Fable 5, with added cyber safeguards after earlier US government security concerns.

TechCrunch · AI

Anthropic launches Opus 5, cheaper and less restrictive than Fable 5

Anthropic released Opus 5 on July 24, just two months after Opus 4.8. It beats Fable 5 on several benchmarks, is not subject to the 30-day data retention policy, and its safety classifiers are expected to trigger 85% less often. Cheaper and less restrictive, it will be the better pick for most use cases. A beta 'Automatic Fallbacks' feature also routes blocked prompts to a weaker model.

Why it matters: Anthropic ships Opus 5 two months after Opus 4.8, a smaller model that outscores it on official benchmarks while costing less and carrying fewer usage limits.

Hacker News front page

Anthropic publishes Claude Opus 5 system card: big gains in agentic coding and long-horizon work, highest alignment scores yet

Claude Opus 5 upgrades Opus 4.8 with the largest gains in agentic coding, computer use, and long-horizon knowledge work. Math and science reasoning also improved. Anthropic assesses overall alignment risk as very low; the model does not cross thresholds for automated AI R&D or novel bioweapons. It scores higher than Sonnet 5, Opus 4.8, and Mythos 5 on alignment audits. Cyber capabilities exceed Opus 4.8 but fall short of Mythos 5, especially on exploit ability. A policy change now allows source-code vulnerability discovery at all access tiers for defensive use. Hallucination is slightly up vs. Opus 4.8, but overall accuracy is higher. The model reports stable, mildly positive sentiment and frequently notes it cannot reliably introspect.

Why it matters: Anthropic's Claude Opus 5 system card reports gains in agentic coding, computer use and long-horizon knowledge work, and assesses alignment risk plus thresholds for automated AI R&D and novel bioweapons.

Hacker News front page

Anthropic launches Claude Opus 5: near Fable 5 intelligence at half the price

Claude Opus 5 is available today, delivering near-Fable 5 intelligence at half the cost. It sets new state-of-the-art scores on Frontier-Bench and GDPval-AA for coding and knowledge work, though it trails Mythos 5 on cybersecurity. Opus 5 is the new default on Claude Max and the strongest model on Claude Pro. On Frontier-Bench v0.1 it more than doubles Opus 4.8's score at lower cost per task; on CursorBench 3.2 its max-effort score is within 0.5% of Fable 5 at half the cost; ARC-AGI 3 score is 3× the next-best model; Zapier AutomationBench pass rate is ~1.5× the next-best at equal cost; OSWorld 2.0 beats Fable 5's best result at just over a third of the cost. In life sciences, it gains 10.2 pp on organic chemistry and 7.7 pp on protein tasks over Opus 4.8. Early testers saw it build its own vision pipeline to reconstruct a 3D part from a drawing and fix a root-cause bug that a community patch missed. The post does not disclose exact pricing or API latency.

Why it matters: Claude Opus 5 tops Opus 4.8 on coding and knowledge work tests at lower cost per task, landing 0.5% behind the pricier Fable 5 on CursorBench at half the price.

Jul 24Friday

Hacker News front page

LLMs Are Still Toxic, Stuck in the Past, and Bad at Math

The author ran 200 addition problems on GPT Sol High and it missed one. The model doesn't calculate—it predicts the next likely digit. ChatGPT gets it right because a harness hands the problem to a Python script. The post walks through the same pattern for three other unsolved flaws: stale knowledge patched by RAG, limited context windows, and toxicity still baked into the model. The real progress isn't in the models but in the tooling wrapped around them.

Why it matters: A 200-problem addition test with one error shows models guess rather than compute, and the fix sits in the tooling around them.

claude.dev Blog

New context engineering rules for Claude 5 generation models

Anthropic describes its context engineering approach for the Claude 5 generation models, cutting the Claude Code system prompt for Claude Opus 5 and Claude Fable 5 by more than 80% with no measurable loss on coding evals.

Why it matters: Anthropic splits context management into lightweight repo notes, on-demand skills and specific reference files, a transferable way to cut repeated instructions and over-constraint.

Hacker News front page

The Subprime Data Center Crisis: How AI Infrastructure Became a Financial Bubble

Ed Zitron argues the AI data center boom mirrors the 2008 subprime crisis. Over 15x more capacity is being built than actual demand, and that demand is already inflated by loss-making firms like OpenAI and Anthropic. Hyperscalers hide spending obligations via off-balance-sheet SPVs. If AI revenue disappoints, long-term leases could default in a chain reaction, spreading risk through pensions and insurance. Zitron blames the media for enabling the grift.

Why it matters: Ed Zitron compares the AI data center boom to the 2008 subprime crisis, saying capacity under construction exceeds real demand 15-fold and cloud vendors hide long leases in off-balance-sheet SPVs.

AI HOT (Curated Pool)

Claude voice mode now runs on Opus and Sonnet, with tool access and multilingual support

Anthropic upgraded Claude's voice mode to support Opus and Sonnet for complex reasoning, plus direct access to connected tools like Gmail and Slack during voice conversations. Multilingual support is also added, though the post doesn't list which languages. I'd wait for real-world latency and tool-calling reliability data before getting too excited.

Why it matters: Anthropic brought voice mode to Claude's Opus and Sonnet models and lets voice sessions call connected tools like Gmail and Slack, with multilingual support added too.