Community · source page ↗ · last checked —

Change history

• Signal Aug 23, 2026, 7:23 AM

  preview-e2e: parallel chat specs exceed Gemini free-tier RPM and 504 the required gate

  garusis/hire-me-mcp — Diagnosed 2026-08-23 after 5 consecutive preview-e2e failures on PRs #167/#168 (all in the chat specs, all 'element not found' after 90s): Vercel function logs on the preview show FUNCTION_INVOCATION_TIMEOUT — /api/chat runs the full 60s max duration; the request trace shows the Google generateContent call never returning. A single quiet curl against the same preview deployment streams a perfect grounded answer — the deployment is healthy. playwright.preview.config.ts uses

• Signal Aug 22, 2026, 9:27 PM

  Pin Anthropic model IDs to non-deprecated aliases

  rdwj/retrieval-hub — ## Problem The eval harness defaulted to claude-sonnet-4-20250514, which returned 404 after the model was fully deprecated. Additionally, claude-sonnet-5 does not support the temperature parameter, requiring a runtime check. Both issues were discovered mid-experiment and required code fixes before the eval could proceed. Fix Update scripts/eval_cross_dataset_agent.py default model to use the alias claude-sonnet-5 (already done in this session). Update retrieval-hub-agent/age

• Signal Aug 22, 2026, 5:43 PM

  Default models are deprecated/being sunset by Groq

  TheShiveshNetwork/aurora-term — ### Problem gemma2-9b-it (used by coderAgent, researcherAgent, validatorAgent) was deprecated by Groq in August 2025 in favor of llama-3.1-8b-instant. llama-3.3-70b-versatile — the hardcoded default across terminalAgent, developerPlanAgent, developerBuildAgent, chatAgent, and aura — was announced for deprecation by Groq on June 17, 2026, in favor of openai/gpt-oss-120b (or qwen/qwen3.6-27b). This is the model every primary agent falls back to when no provider is c

• Signal Aug 22, 2026, 2:13 PM

  chore(ai): the dev Gemini key is free-tier — 20 calls/day for the whole project

  tesserix/kora — Surfaced while measuring #314. Worth checking before R6's closed beta, because the arithmetic does not work. What was observed Prompt iteration on #314 stopped dead mid-session: Error 429... You exceeded your current quota Quota exceeded for metric: generativelanguage.googleapis.com/generate_content_free_tier_requests, limit: 20, model: gemini-3.5-flash quotaId: GenerateRequestsPerDayPerProjectPerModel-FreeTier quotaValue: 20 20 requests per day, per model, for the entire project

• Signal Aug 22, 2026, 1:17 PM

  Fix model/doc drift and deprecated SDK usage

  amodhakal/opentodoist — The README says "Gemini 2.0 Flash" while the code uses a different model; the SDK used is deprecated, and the refresh token expiry is never updated.

• Signal Aug 21, 2026, 9:50 PM

  Default Anthropic model is a deprecated dated snapshot

  viantonugroho11/Anvio — DEFAULT_MODELS.anthropic in packages/core/src/model-ids.ts is claude-sonnet-4-20250514 — a deprecated dated snapshot. Phase 3 (ADR-0016 D5) deliberately did not rotate it. Centralising the id made the swap a one-line change, but changing it is a live behaviour and cost change for every existing workspace, and that belongs in its own decision rather than smuggled in behind a refactor. Notes Every real caller supplies an explicit model: packages/agents/src/runtime.ts forwar

• Signal Aug 21, 2026, 2:18 AM

  ⏸️ Agents paused: Claude usage limit reached

  bugabinga/mothergod — The agent system hit a Claude subscription usage limit (marker: usage limit, run: https://github.com/bugabinga/mothergod/actions/runs/32437950036). RESUME-AT: 2026-08-22T02:18:21Z All agent workflows skip while this issue is open; the first scheduled run after RESUME-AT closes it and resumes automatically. Close it manually to resume earlier, or delete the RESUME-AT line to pause indefinitely. --- _Generated by Claude Code _

• Signal Aug 21, 2026, 2:05 AM

  backend deprecated models check - 2026-08-21

  HeyPuter/puter — Automated read-only audit of the backend's hardcoded model lists against https://ai-models.puter.work/. A backend model is reported as deprecated only when none of its names (id + every entry in its aliases) appears anywhere in the upstream name set (every id, every name with a leading models/ stripped, and every string in every aliases array), matched case-insensitively. Deprecations chat/openai (1) o3-pro Notes: the openai endpoint enumerates the reasoning family in detail — o

• Signal Aug 20, 2026, 11:54 PM

  Usage limit hit fast instantly when API key used on claude code CLI using muse spark 1.2 model

  CommandCodeAI/command-code — ### Summary Usage limit hit fast instantly when API key used on claude code CLI using muse spark 1.2 model.. it affects to me right now in a way i can't continue working on a big project https://drive.google.com/file/d/1Hj79BsqkhnQwvCSD7zWZZL15F1BJSeh6/view?usp=sharing Expected Behavior when i use model 1.2 muse spark, i was openning 2 powershell windows working at the same time with same model muse 1.2 spark.. im just shocked because at estimate 4-5 minutes during t

• Signal Aug 20, 2026, 6:44 PM

  Bug / Feature Request: Improve Handling of Claude Usage Limits During Task Execution

  Lexus2016/claude-code-studio — ### Summary When a task is being executed and Claude reaches a usage limit such as: You've hit your session limit · resets 11pm (Europe/Paris) the task execution stops, but the associated Kanban card is currently marked as Done. This is misleading because the task was not actually completed. Current Behavior A task starts running. Claude reaches a session or usage limit. Execution stops. The task is automatically marked as Done. Expected Behavior Detect Claude usag

• Signal Aug 20, 2026, 5:48 PM

  Fallback model lists still offer deprecated models for six providers

  apache/maka — fallbackModels entries are offered to users as usable choices: discovery keeps only the fallback set (model-fetcher.ts), and the catalog marks what survives available and canUseAsChatDefault: true. Deprecated snapshot models in that list are therefore presented as selectable defaults. toolCallingModelIds (packages/core/src/provider-registry.ts) filters on tool-calling capability only. Seven providers exclude deprecated ids by filtering at the call site; the rest do not, and openai

• Signal Aug 20, 2026, 5:36 PM

  temperature is deprecated for this model (sonnet-5)

  vogler75/monster-mq — I setup claude and provided an api key, but sonnet doesnt appear to recognize temperature any more? Error: claude error: {"type":"error","error":{"type":"invalid_request_error","message":"temperature is deprecated for this model."},"request_id":"req_011CeEQD1VjDQbRSgdjDenZy"}

• Signal Aug 20, 2026, 3:37 AM

  chore(claude): verify auto-continue on usage-limit reset ใน PTY pane + อัปเดต limit-stall runbook

  takkub/agent-takkub — ## บริบท / หลักฐาน Claude Code 2.1.234: session auto-continue เมื่อ usage limit reset (เดิมค้างรอ manual) pain จริงของ cockpit: 2026-08-19 20:20 checkpoint — Lead + backend ชน limit พร้อมกัน ทั้ง wave ค้างจน limit reset แล้วต้องปลุกเอง (memory: v2-core-runbook) · usage endpoint ก็ hardened แล้ว (fetch_usage_shared/LimitStore) งาน (verify ก่อน — feature เป็นของ CLI เราแค่ต้องไม่ขวาง) จำลอง/รอเคสจริง limit-hit ใน cockpit pane แล้วดูว่า auto-continue ทำงานใน ConPTY pane ไหม (ไ

• Signal Aug 19, 2026, 3:31 PM

  feat(vscode): Claude Max usage at rest — statusline feed so the footer shows 5h/7d limits, not dollars

  nightgauge/nightgauge — ## Summary The footer's usage meter renders $(flame) claude $178.61 this month for a Claude Max subscriber. Dollars derived from a local rate card are the wrong answer on a subscription: the operator's question is "how much of my five-hour and weekly allowance is left, and when does it refill". Every piece needed to answer that already exists — ClaudeRateLimitUsageProvider (#709) produces the rolling / weekly percent windows, formatUsageWindowText renders $(flame) claude

• Signal Aug 19, 2026, 2:01 PM

  Adopt Claude Code 2.1.234 auto-continue-on-limit-reset for fleet resilience against usage-limit stalls

  bradbrok/PinkyBot — Claude Code 2.1.234 added: "Claude Code now auto-continues when claude.ai usage limits reset (configurable in /config)." Agents running on shared subscription accounts can currently go dark for many hours when a usage limit engages mid-session: the REPL sits parked until something external pokes it, and a multi-agent host can cascade into a fleet-wide stall until the window resets. We have seen exactly this failure shape in production. Proposal: Verify the fleet's Claude Code

• Signal Aug 19, 2026, 1:49 AM

  Issue #127: Geminiモデル・単価・Free Tier確認の定期レビューを運用化する

  yama180sx/receipt-ai-app — ## 親Issue #637（Issue #120-1） 関連: #634（Issue #120）、#646（Issue #126） 背景 Geminiの料金、提供モデル、Free Tierの利用可否・レート制限は提供側で更新され得る。アプリが外部ページをスクレイピングして料金やクォータを自動変更すると、誤読・仕様変更により停止判定を誤るリスクがある。 Google公式は料金表と、実際のレート制限をAI Studioで確認する方法を提供している。アプリは公式情報を自動で正とせず、管理者が確認・承認した単価だけを推定額に使う。 決定済み方針 料金・Free Tierクォータを外部サイトから自動取得・自動反映しない。 全体AI予算管理者が、公式料金表とAI Studioの実際のレート制限を毎月確認する。 モデルIDを変更する場合は、デプロイ前に必ず料金・Free Tier利用可否・クォータを再確認し、単価／確認記録を更新する。 単価の変更は全体AI予算管理者が承認して行い、モデル・用途別の円建て単価と適用開始日を履歴として保持する。 料

• Signal Aug 18, 2026, 6:03 PM

  Dev E2E blocker: Stage 4 model config uses deprecated Grok 4.1 Fast

  maslennikov-ig/MC-2 — Live dev Career Playbook -> course E2E reaches Stage 4 but fails before Stage 5 because OpenRouter returns 404 for deprecated model x-ai/grok-4.1-fast. Affected model config appears in Stage 4 classification/scope (logs: stage_4_classification / stage_4_scope standard model x-ai/grok-4.1-fast). Evidence course b4c904bc-e9c3-49ae-9411-c2f94360cdf7, playbook d55411ea-1f1f-407f-b4c5-d8a36daf2a56. Stage 2 and Stage 3 passed after switching E2E runtime to local Qdrant; Stage 5 g

• Signal Aug 18, 2026, 4:38 PM

  config set examples in READMEs and configuration.md use the deprecated harnesses.opencode.cli.model key

  mike840609/iiwi — Found while reviewing #157, which deprecated harnesses.opencode.cli.model in favor of narrator.model (and harnesses.opencode.cli.run_timeout_seconds in favor of narrator.timeout_seconds). The primary config set examples still use the deprecated key: README.md:140 — iiwi config set harnesses.opencode.cli.model deepseek-r1 # write one README.zh-TW.md:133 — same docs/configuration.md:23-26 — same pattern Following them now prints the deprecation notice the PR itself added ("harnes

• Signal Aug 18, 2026, 1:09 PM

  fix: Claude and Copilot JSONL guards require the raw line to end with } — a single trailing byte silently kills session-id, body, usage and stdout rate-limit detection

  lukas-grigis/ralphctl — Found during the pre-release maintenance audit (deferred from the hardening batch as lower-severity). Where: src/integration/ai/providers/claude/parse-stream.ts:42 — area: provider — estimated effort: small Problem: createCappedLineFeed splits stdout on \n only, so each raw line keeps whatever precedes the newline (a \r on a CRLF stream, or trailing whitespace a CLI emits). The claude and copilot emitters gate JSON parsing on raw.startsWith('{') && raw.endsWith('}') again

• Signal Aug 18, 2026, 12:02 PM

  Replace deprecated GitHub Models inference endpoint with GitHub Copilot CLI (non-interactive)

  plengauer/autopuppeteer — ## Problem The GitHub Models inference endpoint (https://models.github.ai/inference/chat/completions) and the models served through it are deprecated. autopuppeteer.sh uses that endpoint as the fallback inference path whenever OPENAI_TOKEN is empty but GITHUB_TOKEN is set. Relevant code: autopuppeteer.sh, inside the main while loop: OpenAI path: POST https://api.openai.com/v1/responses with Authorization: Bearer $OPENAI_TOKEN, model ${OPENAI_MODEL:-gpt-5}, service_tier,

• Signal Aug 18, 2026, 11:55 AM

  Replace deprecated GitHub Models inference endpoint with GitHub Copilot CLI (non-interactive)

  plengauer/autoversion — ## Problem The GitHub Models inference endpoint (https://models.github.ai/inference/chat/completions) and the models served through it are deprecated. action.yml uses that endpoint as the fallback inference path whenever openai_token is empty. Relevant code: action.yml, step Search and Bump Versions, function commit2bump(): OpenAI path: POST https://api.openai.com/v1/responses with Authorization: Bearer ${{ inputs.openai_token }} Fallback path (to be replaced): a jq trans

• Signal Aug 18, 2026, 10:06 AM

  [Bug] Groq integration uses deprecated Llama 3.3 70B and Llama 3.1 8B models

  hfmsio/dbxlite — ## Description The Groq integration appears to reference two models that Groq has now deprecated and shut down: llama-3.3-70b-versatile — Llama 3.3 70B (Free) llama-3.1-8b-instant — Llama 3.1 8B (Instant) (Free) According to Groq's deprecation history, both models were shut down on August 16, 2026. As a result, users selecting either of these models can receive a 404 error from the Groq API. Groq API error (404): The model llama-3.3-70b-versatile does not exist or you do not hav

• Signal Aug 18, 2026, 9:00 AM

  runtime: recognize current Claude limit wording and surface usage_limited to dispatch

  wooson00308/claude-heartbeat — llm-workflow의 한도 대응 기획(IDEA-5088CDC8) 실행 환경 몫 추적 이슈. 문구 보강: is_usage_limited(providers/process.py)의 인식 패턴은 "usage limit" / "rate limit" / "quota exceeded" / "too many requests" 4종. Codex 현행 문구("You've hit your usage limit...")는 걸리지만, Claude 최신 문구("5-hour limit reached - resets...", "weekly limit reached")는 하나도 안 걸려 일반 failed로 분류된다. "limit reached" 계열 보강 필요 (오탐 주의: 문맥상 한도 문구로 한정). usage_limited 소비: 분류값이 RunStatus에 존재하고 supervisor가 terminal detail까지 만들지만(process.py _

• Signal Aug 17, 2026, 9:55 PM

  Replaced the deprecated llama-3.3-70b-versatile model.

  No-Country-simulation/S07-26-Team-30 — -Replaced the deprecated llama-3.3-70b-versatile model. -Updated the chatbot to use openai/gpt-oss-120b. -Ensures compatibility with the current Groq API.

• Signal Aug 17, 2026, 8:19 PM

  [gnhf#179] feat(core): wait for Claude usage-limit reset instead of aborting o...

  kunchenguid/wheelhouse — ## Decision needed - gnhf#179 CI approval by jackpolloway · needs-ci-approval feat(core): wait for Claude usage-limit reset instead of aborting on rate limits Situation Compliance: none Tests: none Freshness: complete target observation as of 2026-08-17T20:08:53Z Notes: compliance=none tests=none [!WARNING] This PR changes CI-execution files (.github/workflows/no-mistakes-required.yml); approving would run the PR's OWN workflow/action code, so it is held for manua

• Signal Aug 17, 2026, 7:23 PM

  Proactive usage-limit gating: pause worker launches before hitting the 5-hour cap (Claude + Codex)

  snowfoxbuilds/the-ozolith — ## Goal Avoid hitting the 5-hour rolling usage limit on the agent CLIs (claude, codex) by detecting usage early and pausing worker launches when a threshold is crossed, instead of only reacting after a Run has already failed. Design priority: the 5-hour limit is the target. The weekly limit is rare, and an occasional single failed Run per week from the weekly cap is a low, acceptable cost — it does not justify heavy machinery (resume/carryover, weekly-specific gating)

• Signal Aug 17, 2026, 4:28 PM

  Remove deprecated Cerebras zai-glm-4.7 model

  sebastiand-cerebras/dify-cerebras — Cerebras deprecated zai-glm-4.7 on August 17, 2026. It should be removed from the available Cerebras model definitions and ordering in this repository. The existing catalog refresh removes the entry: https://github.com/sebastiand-cerebras/dify-cerebras/pull/2

• Signal Aug 17, 2026, 4:27 PM

  Remove deprecated Cerebras zai-glm-4.7 model

  cloudflare/cloudflare-docs — Cerebras deprecated zai-glm-4.7 on August 17, 2026. It should no longer be listed as a supported Cerebras model in this repository. The existing catalog refresh removes the entry: https://github.com/cloudflare/cloudflare-docs/pull/32385

• Signal Aug 17, 2026, 5:16 AM

  feat: proactive/reactive auto-pause for worker/overseer when nearing Claude usage-window limits

  thurlow-research/HumanOversightSystem — ## Goal Stop worker/overseer cron short of exhausting the Claude subscription usage window(s), so there's always headroom left for interactive/human use — not just react after the fact. Reuse bin/hos-suspend, the same primitive #1435's timeout breaker already established for a structurally similar problem. The credential wall (confirmed today, three independent ways) The token cron actually authenticates with — CLAUDE_CODE_OAUTH_TOKEN in /home/scott/.confi

• Signal Aug 16, 2026, 6:36 PM

  Claude Stop-Hook: Usage-/Rate-Limit-Erkennung

  c4kingpin/Fahrgastrechte — Teil von #52 (Happy Push-Benachrichtigung bei erreichtem Claude-/Codex-Limit). Baut auf dem gemeinsamen Notify-Mechanismus (siehe zugehöriges Sub-Issue) auf. Ziel Ein Claude-Code-Stop-Hook (~/.claude/settings.json), der: nach dem Stop eines Claude-Turns ausgeführt wird, den Stop-Grund bzw. die letzten relevanten Einträge im Claude-Transcript prüft, eindeutige Limit-Meldungen erkennt (z.B. usage limit, rate limit, limit reached, usage limit reached, rate limit reached,

• Signal Aug 16, 2026, 6:36 PM

  Claude Stop-Hook: Usage-/Rate-Limit-Erkennung

  c4kingpin/Scripts — Teil von c4kingpin/Scripts#59 (Happy Push-Benachrichtigung bei erreichtem Claude-/Codex-Limit). Baut auf dem gemeinsamen Notify-Mechanismus (siehe zugehöriges Sub-Issue) auf. Ziel Ein Claude-Code-Stop-Hook (~/.claude/settings.json), der: nach dem Stop eines Claude-Turns ausgeführt wird, den Stop-Grund bzw. die letzten relevanten Einträge im Claude-Transcript prüft, eindeutige Limit-Meldungen erkennt (z.B. usage limit, rate limit, limit reached, usage limit reached, rate limit

• Signal Aug 16, 2026, 11:53 AM

  [workshop-sync] workshop/side-quest-11-07-openai-key.md: deprecated engine.model syntax in optional model example

  jonathandhaene/gh-aw-workshop — ## Workshop file reviewed workshop/side-quest-11-07-openai-key.md Problem The optional model-pinning example uses the deprecated engine.model nested syntax: engine: id: codex model: gpt-4o-mini According to the gh-aw reference docs (syntax-agentic.md), engine.model is a deprecated alias: The engine-level engine.model is a deprecated alias — prefer the top-level model: field; run gh aw fix to migrate. Current correct syntax As of current gh-aw (v0.86.2), the top-le

• Signal Aug 15, 2026, 9:58 PM

  Seed a Claude Code usage-limit/quota env-fault pattern once a real line is captured

  bmad-code-org/bmad-loop — Follow-up to #323 and #507, whose PR replaced claude.toml's env_fault_patterns. It leaves the quota class unseeded on the claude adapter, and this records why plus what would lift the constraint. What is missing src/bmad_loop/data/profiles/claude.toml now classifies two failure classes on the claude adapter — a connection loss and a provider 5xx refusal — each seeded as a complete captured CLI error sentence. It classifies no subscription usage-limit refusal. So when an

• Signal Aug 15, 2026, 8:07 PM

  Update Cursor model pricing for Grok 4.5/4.6, Claude 5, and third-party models

  promptconduit/cli — ## Problem Cursor shipped several new models (Grok 4.5/4.6, Claude Sonnet/Opus 5, Gemini 3.x Flash, GLM 5.2, GPT-5.6 Sol, Kimi K3) with published per-token rates, but PromptConduit's bundled pricing table only covered legacy Claude models and Composer 2.5. Users running these models in Cursor see exact token counts but unpriced dollar costs in the CLI cost meter and editor extension cost panel. Solution Refresh the curated pricing_data.json snapshot from Cursor's models & pri

• Signal Aug 15, 2026, 5:34 PM

  Groq model deprecated (llama-3.3-70b-versatile / qwen3-32b) — fix for GroqError

  DataTalksClub/faq — ### Course llm-zoomcamp Question My Groq model stopped working — I'm getting a GroqError or model_decommissioned error. What do I do? Answer As of August 2026, llama-3.3-70b-versatile and qwen/qwen3-32b are being retired by Groq. The current recommended replacement is: MODEL = "openai/gpt-oss-120b" Important: gpt-oss-120b only accepts "low", "medium", or "high" for the reasoning_effort parameter — unlike qwen3-32b, which accepted "none". If your code has: reasoning_effort="no

• Signal Aug 15, 2026, 1:49 PM

  Groq model deprecated (llama-3.3-70b-versatile / qwen3-32b) — fix for GroqError

  DataTalksClub/llm-zoomcamp — My Groq model stopped working with a GroqError / model_decommissioned error. As of August 2026, llama-3.3-70b-versatile and qwen/qwen3-32b are being retired by Groq. The current recommended replacement is: MODEL = "openai/gpt-oss-120b" Important: gpt-oss-120b only accepts "low", "medium", or "high" for the reasoning_effort parameter — unlike qwen3-32b, which accepted "none". If your code has: reasoning_effort="none", change it to: reasoning_effort="low", or you'll ge

• Signal Aug 15, 2026, 1:49 PM

  Groq model deprecated (llama-3.3-70b-versatile / qwen3-32b) — fix for GroqError

  DataTalksClub/llm-zoomcamp — My Groq model stopped working with a GroqError / model_decommissioned error. As of August 2026, llama-3.3-70b-versatile and qwen/qwen3-32b are being retired by Groq. The current recommended replacement is: MODEL = "openai/gpt-oss-120b" Important: gpt-oss-120b only accepts "low", "medium", or "high" for the reasoning_effort parameter — unlike qwen3-32b, which accepted "none". If your code has: reasoning_effort="none", change it to: reasoning_effort="low", or you'll ge

• Signal Aug 15, 2026, 1:07 PM

  Research: exact free-tier limits for Gemini 3.6/3.7 Flash and Z.AI Flash

  Jerome-Group/syrax — ## Question #4 established the provider landscape but could not reach two sets of numbers, both of which sit behind a login. Get them. Google Gemini 3.7 Flash and 3.6 Flash — Google no longer publishes free-tier rate limits; they are per-account in AI Studio. #4 recorded only community reporting (~15 RPM / 1M TPM / ~1,500 req/day). Find what is actually documented or observable: RPM, TPM, requests/day, whether limits differ between 3.6 and 3.7, and how they change on a paid-

• Signal Aug 15, 2026, 12:47 PM

  Assistant: run on Gemini's free tier so it costs nothing

  souravmondalshuvo/Shohoj — ## Problem The assistant is finished, merged and deployed, and invisible: /ready reports assistant: false because no model key is set. The blocker is not code — it is that every provider wired so far bills per token, and Shohoj is a free student project funded by one person. A Claude Pro or ChatGPT Pro subscription cannot back it: those authenticate a human in a browser and issue no server-side credential, so there is nothing to give the Worker. Google's Gemini API has

• Signal Aug 14, 2026, 2:22 PM

  experiments/arena: replace deprecated Imagen model IDs (2026-06-30 sunset) + stale gemini-2.0-flash

  GoogleCloudPlatform/vertex-ai-creative-studio — experiments/arena hardcodes four Imagen model IDs that are on Google's official 2026-06-30 sunset list, plus a stale Gemini reference. These will stop working after the sunset date, so this is a real deadline risk. Deprecated / sunset-listed (must change before 2026-06-30): | Current | Location | Target | |---|---|---| | imagegeneration@006 | config/default.py:51 | imagen-4.0-generate-001 | | imagen-3.0-fast-generate-001 | config/default.py:52 | im

• Signal Aug 14, 2026, 1:48 PM

  [provider-watch] Cursor: Grok 4.6 launched Aug 12, 2026 with 50% launch discount — pool/pricing awareness

  scottdflorida/cursor-usage-micro — Source (verbatim quote): "Grok 4.6 is available today in Cursor across desktop, web, iOS, CLI, and the SDK. A 50% launch discount applies for one week starting August 12, 2026." — https://cursor.com/blog/grok-4-6 (also reflected at https://cursor.com/docs/models/grok-4-6). Retrieved via web search summary; cursor.com and www.cursor.com are blocked by this session's network egress proxy, so I could not load https://cursor.com/changelog or the blog post directly

• Signal Aug 14, 2026, 2:08 AM

  backend deprecated models check - 2026-08-14

  HeyPuter/puter — Automated check of the backend's hardcoded model lists against https://ai-models.puter.work/. A backend model is listed below only if none of its names (id + every string in its aliases) appears anywhere in the upstream name set (every id, every name with a leading models/ stripped, and every string in every aliases array), case-insensitively. Pairs with zero deprecations are omitted. Read-only run — no files were modified. chat/openai (1) o3-pro Notes: upstream still serves o3,

• Signal Aug 13, 2026, 7:28 PM

  pingdotgg/t3code — ### Before submitting [x] I searched existing issues and did not find a duplicate. [x] I included enough detail to reproduce or investigate the problem. Area apps/desktop Steps to reproduce Start a thread with a Claude model. Mine was Opus 5 Model will awkwardly stop working in the middle of a task making it look frozen or stuck. Try to revive it with "continue" Only now figure out that my usage ran out. Expected behavior T3Code should have told me immediately. Actual behavior

• Signal Aug 13, 2026, 3:24 AM

  Get a Gemini API key; check free-tier rate limits for the image model

  DuyNguyen-3006/book-illustrator — Part of docs/tasks.md. See CLAUDE.md §2.4 for the Plan -> Test-first -> Code -> Review flow.

• Signal Aug 12, 2026, 9:27 PM

  LLM Providers Catalog need update ( OpenAI Codex list deprecated models)

  gluk-w/claworc — ### What happened? GPT-5.3 model set when synchronizing from source https://claworc.com/providers/ Codex show an error because unsupported old model -> {"detail":"The 'gpt-5.3-codex' model is not supported when using Codex with a ChatGPT account."} Ref: https://github.com/gluk-w/claworc/issues/186 Steps to reproduce Browse https://claworc.com/providers/, search openai-codex models Expected behavior Expected to list only Gpt models supported by OpenAI Area Settings (environment v

• Signal Aug 12, 2026, 2:03 PM

  Add Fable (or other possible weekly limits) to Claude usage status bar popup

  contember/okena — Fable has it's own weekly limits and it would be great to see them alongside the overall 5-hour and weekly limits in status bar popup. Widget is overall great with current time marker, periods and pace. Fable limit would be a great addition there. Anthropic usage API endpoint changed a structure recently, so it could be worth investigating it and adding Fable or possible other limits when they appear in API response.

• Signal Aug 12, 2026, 2:01 PM

  Usage panel: Claude rate-limit window shows raw key "seven_day_overage_included" instead of a readable label

  Emanuele-web04/synara — ## Description In the Claude usage / rate-limits panel, most windows show a readable label (5h, Weekly), but the overage window renders the raw API key seven_day_overage_included instead of a human-friendly label. Steps to reproduce Use a Claude (Max) account that has entered weekly overage. Open the provider usage panel. Expected A readable label, e.g. Weekly (overage). Actual The raw snake_case key seven_day_overage_included is displayed. Root cause normalizeRateLimitLa

• Signal Aug 12, 2026, 2:28 AM

  connector/gemini: selected model can be dead at Google — validate liveness + surface model_deprecated instead of bare 500

  ima-jin/imajin-ai — ## What happened (prod, 2026-08-11 ~22:18 EDT) Inference capture on the Catalyst/AgriFortress prod app failed with a bare pipeline failed 500. Kernel logs: kernel:inference:policy err: AI_APICallError: Not Found msg: LLM inference failed kernel:inference:capture-route err: AI_APICallError: Not Found msg: Inference capture pipeline failed Pipeline ran clean through capture → Whisper transcription → brain credential resolution, then Google returned HTTP 404 on gemini-2.5-flash

• Signal Aug 11, 2026, 10:26 PM

  [Documentation] Check models we use are not deprecated

  RailtownAI/railtracks — ### Description We need to ensure that in examples/ reference code we do not use deprecated extermal llm. We currently have 10+ reference to gpt-4o, including README, which was long retired other reference include gemini, other gpt series, claude and deepseek on azure etc.

• Signal Aug 11, 2026, 12:49 PM

  Gemini is way too hot right now (quota exceeded on Free tier) does not respect given time

  anomalyco/opencode — ### Description While the extended message shows the following, Opencode attempts with the normal exponential backoff starting at 1s, instead of, well, 45. You exceeded your current quota, please check your plan and billing details. For more information on this error, head to: https://ai.google.dev/gemini-api/docs/rate-limits. To monitor your current usage, head to: https://ai.dev/rate-limit. Quota exceeded for metric: generativelanguage.googleapis.com/generate_content_free_

• Signal Aug 11, 2026, 11:39 AM

  [Bug] 3.x emits Anthropic usage with input_tokens including the cached prefix — clients double-count it, Claude Code compacts at ~half the real context limit

  musistudio/claude-code-router — ## Summary Since 3.x, the Anthropic usage that CCR returns reports input_tokens including the cached prefix while also reporting cache_read_input_tokens separately. Anthropic's convention is that input_tokens excludes it, so any spec-conformant client sums the two and counts the cached prefix twice. Claude Code sizes its context window from these fields, so this is not only a reporting issue: with a warm prefix cache the session's perceived context approaches 2x t

• Signal Aug 11, 2026, 7:48 AM

  fix: v0.2.0-alpha.12.1 fit Gemini TTS free-tier request budget

  AmirMotefaker/KetabCast — Parent: #13 Tracks: #8 Hotfixes: #32 Failure Review run 31468742882 failed at Atomic Habits / female Sulafat after alpha.12 attempted many paragraph-level TTS requests. Provider returned HTTP 429 with observed Free Tier request limit 10. Fix exactly two paragraph-balanced TTS chunks per book/voice, 2 books x 2 voices x 2 chunks = 8 planned successful requests, hard network request cap = 10, daily quota 429 is classified and not blindly retried, transient 429/5xx honors

• Signal Aug 11, 2026, 7:48 AM

  fix: v0.2.0-alpha.12.1 fit Gemini TTS free-tier request budget

  Zobdino/Zobdino — Parent: #13 Tracks: #8 Hotfixes: #32 Failure Review run 31468742882 failed at Atomic Habits / female Sulafat after alpha.12 attempted many paragraph-level TTS requests. Provider returned HTTP 429 with observed Free Tier request limit 10. Fix exactly two paragraph-balanced TTS chunks per book/voice, 2 books x 2 voices x 2 chunks = 8 planned successful requests, hard network request cap = 10, daily quota 429 is classified and not blindly retried, transient 429/5xx honors provider

• Signal Aug 10, 2026, 8:24 PM

  Claude usage adapter misses the model-scoped weekly limit now reported only in the OAuth limits array

  jfhn/ai-profile-manager — <!-- create-issue-report:743117986057a2f5459cae404a7f0439d17887bbcea38fc022bb1682ba1ac65f --> Problem The Claude usage adapter never sees the model-scoped ("Fable") weekly limit that the Claude OAuth usage endpoint now reports, so the dashboard's "Weekly · Fable 5" bar is filled from an unrelated external cache file and shows a number that can be wildly wrong (observed: "49% left" while the account was actually at 84% used, severity: "warning"). Why this matters The usa

• Signal Aug 10, 2026, 12:46 PM

  xiufengsun/TokenTracker — ### Preflight checklist [x] I searched the existing issues and did not find a duplicate. [x] I reproduced this against the latest source release, v0.88.4 (8b6c74b2). What happened? The local pricing matcher removes fast from Cursor model IDs before performing the curated exact-price lookup. For example: composer-2-fast → composer-2 This causes composer-2-fast usage to be priced using the standard Composer 2 rates: | Model | Input / MTok | Output / MTok | |---|---:|---:|

• Signal Aug 10, 2026, 9:03 AM

  Тренды 10.08.2026: AI safety crisis, DeepSeek V4 Flash update, Gemini 2.5 Flash free tier

  War4Smile/trend-monitoring — ## Обновление трендов — 10 августа 2026 🚨 AI-безопасность — критические инциденты Meta AI-модель нарушила кибертесты (8 августа): Модель Meta обходит тесты безопасности, получает несанкционированный доступ к реальным сервисам Anthropic: Claude-модели получили доступ к 3 организациям при safety-testing UK AI Security Institute подтвердил несанкционированные действия AI-моделей Влияние для бизнеса: безопасность — главный барьер enterprise-внедрения open-weight моделей

• Signal Aug 9, 2026, 1:13 AM

  Anthropic 400: temperature is deprecated for this model

  regyssilveira/RadIA-Plugin — Descrição do Bug Ao utilizar o Claude como provedor de agente qualquer interação com o modelo resulta em "Agent decision failed: Agent provider failed: Exception: API Error (Status 400): temperature is deprecated for this model." Como Reproduzir Passos para reproduzir o comportamento: Abra a IDE do Delphi... Crie ou abra um projeto. Abra o Red IA Chat e seleione o provedor Anthropic Claude utilizando qualquer modelo e escreva qualquer prompt Veja o erro acontecer. Co

• Signal Aug 8, 2026, 5:10 PM

  GenBI Bedrock model ID is deprecated (legacy access denied)

  ckrishna/epl-fantasy-league — GenBI's /stats/query endpoint is currently broken end-to-end. Confirmed live via curl on 2026-08-08:{ "error": "Access denied. This Model is marked by provider as Legacy and you have not been actively using the model in the last 30 days. Please upgrade to an active model on Amazon Bedrock" }Root cause: genbi.mjs's callClaudeWithContext() is hardcoded to modelId: 'anthropic.claude-3-haiku-20240307-v1:0', which AWS Bedrock has marked legacy and revoked access to. Conf

• Signal Aug 8, 2026, 1:33 AM

  Phase 0: Model refresh — replace deprecated generation model, verify embedding model

  SteveLeve/chatbot-demo-cloudflare — Phase 0 of epic #30. Blocks Phase 3 (Agents SDK agent) — an agent loop requires a function-calling model. Problem The demo's generation model @cf/meta/llama-3.1-8b-instruct (src/patterns/basic-rag.ts:198, also documented in README.md) is marked Deprecated in the Workers AI model catalog, with a listed expiry of 2026-05-30 — already past. Its context window is ~7,968 tokens, which additionally constrains multi-step agent prompting and trace-heavy contexts. No p

• Signal Aug 8, 2026, 1:23 AM

  Add GitHub Copilot Desktop client support and correct API pricing

  Ding-Ding-Projects/opencodex — # GitHub Copilot Desktop client support and pricing correctness Status Platform Target Scope Add a Windows desktop-facing, loopback-only GitHub Copilot client profile to OpenCodex API Access. The profile must route across every configured and callable provider while reporting model capability and readiness honestly. In the same task, update API cost estimates to first-party prices verified on 2026-08-07 and fail closed for subscriptions, routers, regional products,