Cursor fed 835 pages of SQLite docs into its new Harness, hitting 80% sqllogictest pass rate in 4 hours; the old Swarm was halted before hour 2 due to code conflicts. The new system isolates planner and worker roles, uses shared design docs and auto-merge, cutting merge conflicts from 70k to under 1k. A mixed-model setup—Opus 4.8 planning, Composer 2.5 executing—cost $1,339 total, roughly 8× cheaper than GPT-5.5 solo. The public minisqlite repo lacks CLI, C API, and cross-process locking, so it's far from production-ready SQLite. This is a capability demo under ideal conditions—fixed spec, dense feedback—not a daily driver for product teams with shifting requirements.

Menlo Ventures partner Matt Murphy told Equity podcast that Anthropic hit a $47B revenue run rate by May 2026, up from $9B in 2025—growth he hasn't seen in 25 years across internet, mobile, or cloud waves. Menlo led Anthropic's $500M Series D at a $4B pre-revenue valuation. Murphy argues the model was never the real moat; Claude Code, MCP, and Claude Skills turned Anthropic into a platform. He also flagged Lovable and Legora as growing even faster, and pushed back on criticism that Anthropic's Mythos launch was more marketing than safety—though the post doesn't detail his counterarguments.

Codeberg members voted to amend the Terms of Use, banning projects that mostly consist of LLM-generated code without human review. The proposal argues such projects have unclear copyright and lack safeguards. The post doesn't define 'mostly' or specify an enforcement timeline.

Fireworks ran ~1,030 real agent tasks comparing Kimi K3 and Fable 5. Head-to-head on SWE: K3 92.4%, Fable 92.6% — a near tie. Under the hood, K3 excels at symbolic math and dev tooling; Fable wins on web and data viz. Routing between the two lifts accuracy to 93% and cuts cost up to 50x on long agent loops. Oracle routing shows K3 can handle 72–96% of tasks, but a practical router still needs an order of magnitude more routing data.

Poolside released Laguna S 2.1 today, a 118B MoE model with 8B active parameters per token and a 1M-token context window. It scores 70.2% on Terminal-Bench 2.1, beating DeepSeek-V4-Pro Max (64.0%) and Inkling (63.8%), and trailing Tencent Hy3 (295B) by only 1.5 points. On DeepSWE long-horizon tasks it hits 40.4% vs DeepSeek-V4-Pro Max's 9.0%. Poolside says training to launch took under nine weeks and published full eval trajectories. The post doesn't disclose training data cutoff or non-coding performance.

Google DeepMind released Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber. 3.6 Flash is the new workhorse—better at coding and multimodal tasks, with up to 17% lower token usage and a cheaper price than 3.5 Flash. 3.5 Flash-Lite targets extreme cost efficiency, and 3.5 Flash Cyber is fine-tuned for finding and fixing security vulnerabilities, available only to governments and trusted partners in a limited pilot. The whole drop is about efficiency, latency, and reliability for customers building AI agents at scale. The real story is what’s absent: the flagship Gemini 3.5 Pro hasn’t been updated since February, while OpenAI shipped GPT-5.5 and started rolling out GPT-5.6, and Anthropic launched Claude Opus 4.8. The post doesn’t explain what’s holding up the Pro line.

Josh Bleecher Snyder now says calling Claude a compiler is a category error — it's better. A compiler only handles source-to-binary decisions, but Claude works vertically across strategy, product, architecture, and machine code. He walks through building exe.dev's distributed DNS server: multiple concurrent agent loops implemented the same design yet made wildly different choices on decisions they never asked about. By answering their questions and reverting bad calls, he slowly converted hard-won knowledge into terse written guidance.

Moonshot AI's Kimi K3 beat Fable and GPT-5.6-Sol on Arena's frontend coding leaderboard and came close on other benchmarks. It's a 2.8T-parameter model with a 1M-token context window and image support; weights will be open-sourced by July 27. Token inefficiency cancels its per-token price advantage: half the cost per token but twice the tokens used. New subscriptions are paused due to GPU shortages. Fable 5 is now a permanent part of Claude Max/Team plans, with Pro users getting a one-time $100 credit. Fable also found a counterexample disproving the 87-year-old Jacobian conjecture. Sierra launched Horizon, outcome-priced long-running agents. NotebookLM rebranded to Gemini Notebook and added Collections.

Anthropic's Cat Wu and Thariq Shihipar shared Claude Code team practices at AI Engineer World's Fair with Simon Willison. Claude Tag, their Slack collaboration tool, now lands 65% of the team's product engineering PRs. They found that stuffing system prompts with examples or 'don't do X' rules hurts output quality on Fable 5 and Opus 4.8, so the Claude Code system prompt shrank by 80%. Thariq noted Fable can edit video to meet their brand team's bar, and the team now treats rewrites as a valid move—the Bun-in-Rust Claude Code already shipped to everyone. The transcript cuts off before detailing what non-engineers do with Claude Tag.

Moonshot's Kimi K3 and Alibaba's Qwen 3.8 launched this week, both near Anthropic Fable 5 in performance and set to release weights publicly. The piece runs the numbers: Anthropic leases data centers and buys electricity, so inference costs scale with usage. Fable 5 costs nearly 3× per completed task vs. competitors. Open models catching up makes a premium-pricing strategy fragile. Anthropic bets on regulation and recursive self-improvement, but its product moat is thin—open-source harness startups are flooding in. The post doesn't spell out a clear countermove.

The author asked Kimi K3 to optimize a hand-tuned Rust function. The model chose a SWAR approach, treating a 64-bit value as eight u8 values, completed the task in about 15 minutes with zero errors, and delivered a 2% speedup. Token cost was roughly $1 via OpenRouter. The author sees this as near-frontier performance.

The UK's AISI found that GLM-5.2 and DeepSeek V4-Pro now trail closed frontier models by only 4–7 months on cyber tasks, down from 6–10 months in 2025. GLM-5.2 matches Claude Opus 4.6 on 70 narrow evals but falls further behind on long-horizon hacking ranges. Kimi released K3, a 2.8T-parameter model that scores near Claude Fable 5 and GPT 5.6 Sol, though the post hints at benchmark overfitting. K3 also wrote a GPU compiler and designed a chip in 48 hours; weights will be released in weeks. DeepMind's Demis Hassabis proposed a FINRA-style US standards body to test frontier models for national security risks, starting with voluntary 30-day pre-release reviews before moving to law.

Cursor redesigned its agent swarm into a tree-shaped planner-worker split and retested it on rebuilding SQLite in Rust from docs. The new swarm beat the old one in every model config: with Grok 4.5 it hit 80% on a held-out SQL test suite in 4 hours, while the old swarm spiraled before hour two. Quality stayed similar across model mixes, but costs varied enormously—the post shows a comparison chart without exact dollar figures. The design isolates context so planners never see implementation details and workers only focus on narrow tasks, preventing drift on long runs.

Approval gates only decide whether a command starts, not what happens after package managers load scripts and spawn child processes. The article walks through a bug-fix task to show how OS isolation (Seatbelt/bubblewrap), credential proxying (Docker Sandboxes), and isolated workspaces each address different risks. No performance numbers or latency figures are disclosed.

Moonshot AI plans to IPO within six months, riding the momentum of its new Kimi K2 model. K2 matches OpenAI o3 and DeepSeek V4 Pro on math and coding benchmarks. The company is valued at about $3 billion, with roughly $150 million in 2025 revenue from Kimi chatbot subscriptions and API fees. The post doesn't specify the listing venue or underwriters. The six-month timeline hinges on market conditions and regulatory approvals—don't bank on it yet.

xAI released the Rust client harness that handles local files, commands, and permissions for Grok Build under Apache-2.0. The Grok model, cloud services, and the official binary build chain remain closed. The repo doesn't accept external PRs. The commit from the earlier upload controversy isn't in the public history, so the current code can't close that case. The real win: you can now pin a public commit, build it yourself, and compare its behavior against the official binary.

The author ran Kimi K3 alongside Claude for coding and couldn't tell them apart on output quality or token usage. K3's API costs $3/$15 per million input/output tokens vs Claude's $10/$50. Subscriptions are even more lopsided: Kimi's $39 coding tier is far more generous, while Claude's $20 plan quietly dropped Fable access because the economics didn't work. The bigger story is US AI policy failure—restricting American models only constrains American customers, while frontier-quality Chinese models like K3 and GLM 5.2 ship without those limits. Semgrep found GLM 5.2 beating Claude on cyber benchmarks precisely because the restricted model declines work the open one just does. The author expects the US to repeat its auto-industry playbook: subsidies and tariffs propping up domestic models that can't compete internationally.

The author tested Claude Fable 5 and GPT-5.6 Sol on an unpublished fiber-network optimization problem, each with three 30-minute runs, comparing plain mode against /goal. Fable 5's plain mean was 32,386—1,875 points lower than Sol's 34,261—and its three plain runs stayed within a 319-point range, showing remarkable consistency. /goal won four of six trials but made both models' means worse: Fable 5 by 759 points, Sol by 868. The feature occasionally gives a small edge but can also cause large regressions. The post also breaks down how /goal differs under the hood: Claude Code uses Haiku as a transcript-only evaluator, while Codex has persisted state and lifecycle tools. Bottom line: Fable 5 is the real story here; /goal is not a safe default.

This piece reframes DSLs as building a smaller executable world for LLMs, not just inventing syntax. The author walks through his own abandoned DSL project and argues that AI has lowered the cost of designing languages, writing parsers, and preparing examples—so teams can build domain boundaries first and decide their lifespan later. The core stack: FastAPI provides atomic actions, DSL describes complete plans, and Context Infrastructure stores past judgments and corrections. The article cites Unmesh Joshi's DSL post and the Tickloom demo, but notes Tickloom is an engineering showcase without cross-model benchmarks proving DSLs improve LLM accuracy. It warns that overly narrow boundaries can exclude correct answers—effect surface and composition rules matter more than Turing completeness.

Cursor's eval lead Nate Schmidt explains on Anthropic's blog how they determined Claude Fable 5 was ready for the hardest 1% of real-world coding problems. The headline number is 72.9% on CursorBench, a significant jump over the prior generation. The post stresses this isn't a generic benchmark grind—it targets long-tail tasks that actually stump developers. The article doesn't disclose the baseline score, test set size, or sample problems, so treat the 72.9% as a directional signal rather than a cross-benchmark comparison point.

Google DeepMind released Gemini 3.5 Flash Cyber, fine-tuned from 3.5 Flash to find, verify and patch vulnerabilities quickly. With multiple calls, it approaches larger models on benchmarks such as CyberGym.

Olaf Alders details how Claude Code v2.1.198 introduced a 60-second timeout that lets the agent proceed without human input—shipped with no changelog entry and no documented off switch. He used Claude itself to reverse-engineer the minified JS bundle and confirmed the logic was buried with no standalone feature flag. Anthropic shipped a fix two days later, but the incident shows Claude Code's auto-update can silently push surprising defaults, and users have almost no visibility into what changed.

Moonshot AI released Kimi K3, a 2.8T-parameter model and the most expensive from a Chinese lab so far at $3/$15 per million input/output tokens—matching Claude Sonnet pricing. Self-reported benchmarks mostly beat Claude Opus 4.8 and GPT-5.5 but lose to Claude Fable 5 and GPT-5.6 Sol. On Artificial Analysis's private long-horizon knowledge eval, K3 hit an Elo of 1547, +732 over K2.6, behind only Fable 5. Cost per task is $0.94, close to GPT-5.6 Sol's $1.04 and roughly half of Opus 4.8. Output tokens dropped 21% vs K2.6. The model only offers a 'max' reasoning effort; Simon's pelican-on-a-bike SVG cost 25 cents and burned 13,241 reasoning tokens. Input token count suggests an ~85-token hidden system prompt. Vision works well. Open weights promised by July 27.

Kimi K3 hit #1 on Frontend Code Arena with 1679 points, beating Claude Fable 5's 1631 and taking six of seven frontend domains. It packs 2.8T params, 1M context, $3/$15 per million tokens, with full weights opening by July 27. Early testers got mixed results: one user's 199-yuan monthly plan produced stunning particle VJ effects from chat history, while another burned through a $40 coding plan in five hours as the model looped on a domain spelling error. Benchmark trust is shaky—GLM-5.2 scored well on paper but felt worse than 5.5 in practice. Writing style drew split reactions: less AI flavor but forced casual tone, nowhere near the natural Chinese of the old Opus 4.6. Same day, GPT-5.6's frontend taste was called 'very Claude-like,' Sol traced a deadlock only reproducible on Ubuntu, Linus told kernel devs AI is here to stay, and Schema harness pushed ARC-AGI-3 efficiency to 98.98% by making models think like physicists.

OpenAI published a CFO-oriented guide that shifts AI spend measurement from cost per token to cost per successful task. It builds a four-question framework: how much useful work gets done, what a successful task actually costs, how dependable the result is, and whether each dollar buys more work as usage scales. GPT-5.6's three tiers—Sol, Terra, Luna—are used as examples; Sol hits 72.7% on DeepSWE v1.1 vs. Claude Fable 5's 69.9%, with 36.2% lower estimated API cost. The post does not disclose specific pricing.

Moonshot AI released its next-gen Kimi model on July 17, matching or nearing OpenAI o3 and Anthropic Claude Sonnet 4.5 on benchmarks like MATH and HumanEval. The model is available for testing via the Kimi chatbot, with an API already live. Nvidia dropped over 3% premarket on the news, as markets worry about pricing pressure on US AI firms. The post doesn't disclose parameter count, training cost, or inference latency, so I'd hold off on those practical metrics.

Kimi K3 scored 1679 on Frontend Code Arena, taking first in 6 of 7 frontend sub-tasks and beating Claude Fable 5 and GPT-5.6 Sol. It's a 2.8-trillion-parameter MoE model with a 1M context window, and open weights are promised for July 27. API pricing is $15 per million tokens—no low-cost play here, it's priced against top closed-source models and aimed at long-context coding and agent workflows.

Moonshot AI launched Kimi K3, a 2.8T total-parameter model and the largest open-weight release to date. Weights are promised by July 27. It supports 1M-token context and native multimodal input, and is live on Kimi.com, Kimi Code, and the API. Artificial Analysis scored it 57 on the AA Intelligence Index—comparable to Claude Opus 4.8 and GPT-5.5, but still behind Claude Fable 5 and GPT-5.6 Sol. On LMSYS Frontend Code Arena, K3 hit #1 with 1679 points and a 76% win rate, well above Fable 5's 63%. Text Arena placed it at #9, a big jump from K2.6's #38. Moonshot itself flagged a noticeable UX gap versus Fable 5 and GPT-5.6 Sol. AA measured cost at $0.94 per task, with 21% fewer output tokens than K2.6.

Laura Summers of Pydantic describes the real fatigue of coding with LLMs: code generation works, but human energy gets drained by constant reviewing, correcting, and holding intent. Her colleague Douwe wakes up to 30 AI-generated PRs daily and feels the pull to delegate review to AI too—raising the question of what he's still doing there. She calls this supervision fatigue and notes AI increases how many things you can start, but not how many you can thoughtfully finish. A Berkeley Haas study is cited: AI doesn't reduce work, it intensifies it.

BackendForge starts with 7,250 tests across 56 backend tasks, then has agents hunt for gaps and add 640 more. Only +8.8% more tests, yet GPT-5.5's passing tasks drop from 31 to 16, and Claude Opus 4.7 from 33 to 10. Same code, sharper questions on permissions, dirty data, and cascading effects. The paper says materials will be released, but they aren't public yet—don't generalize these numbers yet.

Soofi S is a 30B open model trained entirely on Deutsche Telekom's Munich cloud by a German AI consortium. It uses a hybrid Mamba-Transformer MoE design, activating only 3.2B parameters per token, so throughput stays nearly flat even at 256K context. The training mix deliberately favors German, and it beats Olmo 3 32B and Apertus 70B on German, English, and coding benchmarks. Critics called it overtrained under Chinchilla scaling laws; the project's tech lead counters that those laws don't hold for MoE and notes Nvidia trained on up to 25T tokens.

Anthropic shared their playbook for large-scale code migrations with Claude Code. The headline case: porting Bun's 1M+ lines of Zig to Rust in two weeks. The approach splits work into planning, execution, and verification — Claude Code reads the codebase, writes a migration plan, generates PRs, and passes CI. Full workflow and prompt templates are included, aimed at teams running AI-assisted refactors internally.

At Local-First Conf in Berlin, the author noticed a shared dissonance: speakers criticized LLMs while the audience applauded with Claude Code open. He concedes every critique—slop, trust erosion in OSS, broken junior-senior teaching loops, geopolitical supply risks—yet still uses LLMs heavily. The post doesn't resolve the tension; it lays out the contradiction and asks others to share their usage patterns so the community can better understand this collective unease.

After 487 days, Roc's 300K-line Rust compiler rewrite in Zig reached feature parity. A demo game now compiles to a 31KB wasm binary, less than half the original size. The new compiler adds hot code loading during dev and reproducible cross-compilation. The team says this was a redesign, not a direct port, so direct Rust-vs-Zig comparisons don't apply. No formal release yet; v0.1.0 is planned later this year.

Thinky released Inkling, a 975B-total, 41B-active MoE model that handles text, image, audio, and video with a 1M-token context window. Trained on 45T tokens and licensed Apache 2.0, it landed with day-0 support from vLLM, Hugging Face, and others. The team frames it as a customizable base for future iterations, not a benchmark-chasing flagship. A 12B-active Inkling-Small preview also dropped. Independent reviewers call it the strongest US open-weight model so far, though it still trails top Chinese open and best closed models on some benchmarks.

xAI released the full Grok Build codebase on GitHub, covering the agent loop, tool dispatch, terminal UI, and extension system. You can read the source to see how context assembly and tool calls work, or compile it yourself and point it at a local inference setup.

bitjaru open-sourced StyleSeed, a design-rules engine for AI coding tools like Claude Code, Codex, and Cursor. It teaches design judgment rather than just generating code: 74 rules, 48 components, 7 brand skins (Toss, Stripe, Linear, Notion, Raycast, Arc, Vercel), a named motion system, and 15 /ss-* skills. MIT licensed, currently at 731 stars. The post doesn't detail how rules are enforced or how the motion system works in practice, but the structure aims to suppress the 'AI-generated' look in shipped UI.

Unmesh Joshi argues on Martin Fowler's blog that upfront specs are just hypotheses—design is discovered through implementation. DSLs and domain abstractions act as a harness for LLMs, setting clear boundaries so generated code matches intent. The Tickloom example shows a domain model for distributed systems where an LLM helps iteratively build the DSL and then serves as a natural-language interface to it. The DSL becomes the source of truth for the system.

On June 5, OpenAI merged PR #26210, encrypting task messages that Codex's parent agent sends to sub-agents. Previously, local session logs showed plaintext instructions like 'Review the authentication changes'; now only <ciphertext> remains. Sub-agent tool calls, commands, and outputs are still visible, but debugging can't tell whether the parent gave a wrong task or the sub-agent misunderstood. Encryption happens server-side in the Responses API; the local client only forwards ciphertext. This differs from earlier hidden reasoning and compaction—what's now hidden is content that directs another agent to act, not internal model thinking. The post doesn't spell out OpenAI's rationale; speculation includes prompt protection or unified cloud multi-agent services.

Dan Austin open-sourced a full AI-trains-AI loop. An outer Qwen3.6 agent designs post-training recipes; an inner Qwen3-0.6B or 1.7B model runs real GPU training, and hidden eval scores feed back as reward to update the outer policy. Over 54 steps, the agent first learned to reduce invalid submissions, then shifted 1.7B model usage from 42% to 95% and began tuning temperature, optimizer, and other hyperparameters. Trained small-model scores rose from noise level into the 0.22–0.48 range, with limited transfer to a held-out triage task. A postmortem also revealed an evaluator bug: the old tool-use detector looked for `.function.name` instead of `.name`, so the 0.4-weighted score never fired—yet the reward curve still climbed. The fix required a full restart. The experiment shows outer RL can reshape a training agent's behavior, but tasks, rewards, and budgets are still human-designed.