— / 05
Sources
Source fileLessons
Every source used in this survey, with what I took from it. All accessed on 13 Oct 2026 unless the note says otherwise.
| Source | What I took |
|---|---|
| 36kr MiMo price cut | up to 99 % cut on 27 May 2026 |
| AA 0731 at v4.1 | 0731 = 50 on v4.1, preview 40 on v4.1 |
| AA changelog | v4.2 / v4.3 / v4.3.2 dates, TB4.0 and AutomationBench added |
| AA Claude Haiku 5.5 | 43 max, 440M tokens, TTFT 303 s |
| AA Endpoint Accuracy Index | method; price does not predict accuracy |
| AA Gemini 3.8 Flash | AA 41 / 40 / 33, TTFT about 25 s |
| AA GLM-5.2 providers | DeepInfra 73, Fireworks 100, Parasail 98 |
| AA GLM-5.3 providers | DeepInfra 96, Parasail 99, Together 97 |
| AA GLM-5.3-Flash | AA 42, 320B / 18B |
| AA GLM-5.3 | AA 45 max / 34 low, 210M tokens, 64 tok/s |
| AA gpt-oss-120b providers | DeepInfra 85 / Turbo 98, Parasail 97, Together 81 |
| AA Kimi K3 providers | Fireworks 97, Parasail 102 |
| AA Kimi K3 | K3 AA 44, output price band |
| AA leaderboard | AA v4.3.2 scores |
| AA Ling 3.1 Flash | AA 41 |
| AA methodology | v4.3.2 composition and weights |
| AA MiMo-V2.6-Flash | AA 38 |
| AA MiMo-V2.6-Pro | AA 46, speed |
| AA Muse Spark 1.3 | AA 48 |
| AA Qwen3.8-Flash-Next | AA 40, verbosity |
| AA V4 launch at v4.0 | preview = 47 on v4.0, hallucination rate 96 % |
| AA V4-Flash 0420 | preview = 24 |
| AA V4-Flash 0731 | bar = 34, 284B/13B, MIT, deprecated |
| AA V4-Pro providers | Endpoint Accuracy: DeepSeek 107, hosts 97–107 |
| AA V4-Pro | AA 36 |
| AA V4.1-Flash | AA 39 / 25, 190 tok/s, TTFT 0.96 s, 250M tokens |
| agentdeals Mistral | business-only API since 5 Aug 2026, free credit unstated since 7 Oct 2026, training opt-out |
| agentdeals.dev: Ollama Cloud | pricing change dates (late Aug to 1 Sep 2026) |
| agentjournal.dev: Vercel free | card or purchase stops free credits |
| aifreeapi EEA note | EEA paid-only note, per-project limits |
| Alibaba Coding Plan | Pro USD 50, request quotas, old models, no-scripts scope |
| Alibaba Token Plan landing page | credits per tier, conversion not stated |
| Alibaba Token Plan | tiers and supported tools |
| Antigravity credits | about 20 requests/day, weekly refresh |
| Antigravity review | suspensions for using quota through other tools |
| apiyi Pro removal | Pro removed from free 1 Apr 2026 |
| Auto Exacto | default for tool calls, re-ranking, tool-error reductions |
| awesome-free-ai-coding: Vercel | zero-priced routes, "assume they train" quote |
| BFCL | frozen since 12 Apr 2026 |
| build.nvidia.com DeepSeek V4.1-Flash | V4.1-Flash on build since 18 Sep 2026, 1M ctx, FP4 KV, effort 1 to 100, trial ToS |
| Claude pricing | Haiku 5.5 price tiers at 100K |
| codingplan.org GLM | plan prices 18 / 80 / 168 |
| ComplianceHub on Garante | Italy blocked DeepSeek |
| computingforgeeks GPT-6 Luna | Luna tools-only-at-effort-none, hallucination 93 % |
| computingforgeeks MiMo test | DevOps hands-on results (K8s 1/3 and 0/3) |
| DeepInfra Anthropic integration | Anthropic-compatible base URL |
| DeepInfra data privacy | not stored to disk, metadata-only logging, no training |
| DeepInfra models API | live IDs, prices, cached prices |
| DeepInfra pricing | DeepInfra list prices |
| DeepInfra rate limits | 200 concurrent requests per model |
| DeepSeek changelog | 2026 release and retirement dates |
| DeepSeek pricing | model IDs, legacy routing, peak/off-peak prices and hours, 384K output, features |
| DeepSeek rate limit | 2,500 / 500 concurrency, 429, keep-alive, 10-minute timeout |
| DeepSeek status | 90-day uptime 99.69 % / 99.92 % |
| DeepSeek-V4 paper | 0420 preview vendor numbers, non-think vs think gap |
| DeepSeek-V4.1-Flash paper | effort 1–100 and 60–80 curve, DeepSWE harness gap (vendor) |
| dev.to measured limits | measured 5 RPM / 20 RPD, 15 / 500 Flash-Lite, latency |
| dev.to NVIDIA free test | SMS verification, ~40 RPM, Kimi K3 reasoning_content echo (1 Oct 2026) |
| eesel MiMo pricing | PAYG prices for Pro, Flash, ultraspeed |
| eesel Step 5 | Step 5 specs |
| Exacto | :exacto, +10–20% on Tau2Bench/LiveMCPBench |
| Exo Free promo | exo-free added 6 Oct 2026 |
| Fire Pass docs | closed to new subscriptions, Kimi K3 Fast |
| Fireworks changelog | 26 Sep removals, 1 Oct price rise, TPM tiers and 7 Oct doubling |
| Fireworks data residency | global routing vs us.api.fireworks.ai |
| Fireworks no-card limits | USD 1 credit, ~10 RPM without card |
| Fireworks pricing | Fireworks prices |
| Fireworks serverless pricing | Standard prices, US +50%, Priority +25%, Batch −50% |
| Fireworks ZDR write-up | ZDR default, Responses API 30-day store, batch exception |
| Gemini 3.8 Flash | 1,048,576 in / 65,536 out, capabilities |
| Gemini API pricing | free models, paid and Flex prices, 2027 doubling |
| Gemini API terms | EEA data clause and paid-only clause, verbatim |
| Gemini CLI quota | replaced by Antigravity CLI 18 Jun 2026 |
| Gemini forum 20 RPD | about 20 RPD on 3.8 Flash |
| Gemini models | model list, 2.5 limited access |
| Gemini rate limits | free limits unpublished, "not guaranteed" |
| GLM Coding Plan FAQ | "strictly limited", no refunds, coding endpoints, error 1113 |
| GLM Coding Plan overview | credits, auto-routing of old models |
| GLM Coding Plan usage policy | supported-tools-only scope, bans |
| HF GLM-5.3 | GLM-5.3 vendor numbers |
| HF Kimi K3 | K3 vendor tool-use numbers, message echo rule |
| HF V4-Flash-0731 card | no Jinja chat template |
| inventivehq Qwen Code | 1,000 → 100 → 0 RPD, 15 Apr 2026 |
| itsfree.ai: Vercel AI Gateway | free-tier-eligible model list (2 Sep 2026), "no card" claim |
| K2 Vendor Verifier | Fireworks and DeepInfra 100% schema accuracy |
| Kilo Gateway docs | base URL, 200 requests per hour per IP, Anaconda acquisition note |
| Kilo Gateway models | Step 5 Preview and Ling free, may-train flags |
| Kilo live model catalog | 393 models, free ids, mayTrainOnYourPrompts flags, kilo-auto/free set |
| Kilo models and providers | data warnings for free and NVIDIA routes |
| Kilo models API | Step 5 free on Kilo, tools, reasoning levels, may-train flag |
| Kimi Code docs | plan scope, User-Agent rule |
| Kimi K3 model card | tool-call message rules |
| Kimi platform pricing | K3 USD 3 / 15 |
| Kimi Vendor Verifier | Fireworks K3 within noise |
| kingy.ai MiMo setup | HTTP 400 without echoed reasoning content |
| klymentiev.com NIM credits | 1,000 credits claim, 40 to 200 RPM claim (2 Oct 2026) |
| Kong DashScope guide | Anthropic-compatible URL |
| llm-stats Qwen3.8-Flash-Next | specs, vendor scores, ID caveat |
| Lyceum roster | Lyceum EU models and prices, zero-retention claim |
| manifest.build 429 FreeUsageLimitError | free-limit error, CLI vs direct API 429 reports |
| Meetily on DeepSeek | training by default with opt-out, no DPA/SCCs |
| MiMo Token Plan | plan prices, credits, off-peak, endpoints, own-backend prohibition |
| Mistral changelog | Large 4 preview 6 Oct 2026, launch discount |
| Mistral GLM-5.3 | Mistral-hosted GLM-5.3 price and context |
| Mistral Large 4 pricing | sale/list price, measured 524K context, licence status |
| Mistral Large 4 specs | 1M claimed, 524k measured context |
| Mistral regional endpoints | EU endpoint +10% |
| Model Studio free quota | 1M tokens per model, 90 days, Singapore only, Free Quota Only switch |
| Model Studio models | intl model list |
| Model Studio pricing | intl PAYG prices incl. DeepSeek idle/busy, batch −50 % |
| Model Studio privacy notice | no training on data |
| Muse Spark Contributor tier | Contributor price, training on data, 60 RPM |
| Muse Spark pricing | Contributor tier cross-check |
| NIM LLM API reference | ids listed in docs but not live (churn) |
| NIM product docs | prototyping-only Developer Program, AI Enterprise USD 4,500 per GPU per year, 90-day trial |
| NVIDIA forum: is NIM free | "unlimited but RPM", not for production (May 2026) |
| NVIDIA forum: new models time out | GLM 5.1 and Kimi K2.6 timeouts (May 2026) |
| NVIDIA live model list | 80 live ids on 13 Oct 2026, strongest free models, ids missing vs docs |
| Ollama Cloud docs | native API, OpenAI / Anthropic subset, no-logging statement |
| Ollama Cloud pricing | Pro USD 20 with USD 60 credits, K3 and GLM-5.3 at list |
| Ollama pricing | plans, credits, concurrency, per-token rates, off-peak window |
| openbenchmarks same-model accuracy | GLM-5.3-Flash tool-task success per host |
| OpenCode Zen docs | paid prices, endpoints, US / EU hosting, docs-only IDs |
| OpenCode Zen models API | live free IDs and data notes |
| OpenRouter BYOK | USD 25,000/month fee-free, then 5% |
| OpenRouter data collection | no prompt storage by default, 1% logging discount |
| OpenRouter endpoints API | per-host prices, quantization, context, max output, 1-day uptime for MiMo, GLM, DeepSeek on OpenRouter, DeepInfra, Fireworks, Parasail |
| OpenRouter free-model tracker | Ling 3.1 Flash added 3 Oct 2026, free variant gone by 13 Oct 2026 |
| OpenRouter limits | 20 RPM, 50 / 1,000 RPD, USD 10 threshold, 402 on negative balance |
| OpenRouter live catalog | Step 5 Preview USD 1.00 / 2.70, Nemotron 3 Ultra free uptime 77.4 %, free list count |
| OpenRouter models API | free list (16 entries, none above bar), :batch prices, dated-ID aliases |
| OpenRouter pricing | 5.5% / 8% fees, EU routing on Business/Enterprise |
| OpenRouter provider logging | separate paid/free training toggles |
| OpenRouter provider routing | routing parameters, :nitro, :floor |
| openrouter-free-models tracker | dated free-model removals |
| OR free-model tracker | dated free-model removals |
| OrcaRouter: Step 5 open weights | Step 5 specs, weights due 15 Oct 2026 |
| OVHcloud capabilities | 2 / 400 RPM, no storage, Gravelines |
| OVHcloud catalog | OVH models and EUR prices |
| Parasail pricing | Parasail prices, "no minimums or rate limits" |
| Qwen Code issue 1742 | free OAuth tier shutdown |
| Qwen3.8-Flash-Next | specs, vendor numbers, model ID caveat |
| Requesty free models | free model IDs and data metadata |
| Requesty models API | DeepInfra :flex prices |
| Requesty pricing | Free 200 RPD, no card, 5% PAYG markup, EU region on all plans |
| Requesty quickstart | OpenAI, EU and Anthropic base URLs, model naming |
| Scaleway pricing | EUR prices, batch -50%, 1M free tokens |
| Scaleway rate limits | base vs Enhanced tier, console quotas |
| Scaleway signup | 1M free tokens for new users |
| Scaleway supported models | IDs, 256k caps, quantization, devstral deprecation |
| Standard Compute NVIDIA | trial ToS data-use summary |
| Standard Compute on Z.ai | DPA: no training unless opt-in, not stored |
| Step 5 open weights | weights date, vendor benchmarks |
| Step 5 Preview | Step 5 specs, weights date |
| Step Plan prices | plan USD prices (third-party) |
| Step Plan | plan credits and models |
| StepFun launch post | "works with your own stack" quote |
| StepFun pricing (CN) | CNY prices |
| StepFun pricing (intl) | Step 5 / 3.7 / 3.5 Flash prices, rate tiers V0 to V4 |
| StepFun pricing | Step 5 Preview USD 1.00 / 2.70, 64K output |
| tbench.ai leaderboard | official TB4.0 cross-check numbers |
| Together changelog | silent redirects on deprecated IDs |
| Together pricing | Together prices for the alternative note |
| Together serverless models | Together quantization disclosure |
| Together ZDR | stored by default, ZDR toggle |
| toolnavs free GLM-4.7-Flash | 1 concurrency, about 1 RPS, about 1,000 RPD |
| TrueFoundry on OR free models | free-model churn, 75-day median, failed requests count |
| TrueFoundry: OpenRouter free models | median free-model listing life ~75 days |
| TypingMind MiMo guide | base URL, 128K output, card required |
| usagepricing Zhipu price change | conflicting 18 / 72 / 160 |
| Vals SWE-bench | SWE-bench Verified scores, archived 1 Sep 2026 |
| Vals TB2.1 | V4.1-Flash 74.5, 0731 67.0, MiMo-Pro 67.8, MiMo-Flash 76.4 |
| Vals TB4.0 | bar 18.7, GLM-5.3-Flash 37.4, V4-Pro 14.1, MiMo-Pro 31.3, MiMo-Flash 24.2 |
| Vellum on Gemini 3.8 Flash | Gemini 3.8 Flash vendor numbers, 2027 price rise |
| Vercel AI Gateway pricing | no markup, USD 5 per 30 days, purchase ends free tier, ZDR pricing |
| Vercel AI Gateway rate limits | free limits unpublished, 429 body, retry-after |
| Vercel Anthropic Messages API | Anthropic-compatible base URL and env vars |
| Vercel community: card verification | 403 customer_verification_required (9 Jul 2026) |
| Vercel live model catalog | 463 models, list prices, zdr / no_training / regions, deprecated_at, V4.1-Flash 32k output |
| vibecoding.app plan review | Coding Plan ban risk |
| Z.ai GLM-5.3 guide | 1M context, 128K output, always-on thinking, effort default max, endpoints |
| Z.ai pricing | GLM-5.3 first-party 1.40 / 4.40, 128K max output |
| Z.ai release notes | GLM-5.3 and GLM-5.3-Flash release dates |
| zentor.ai GLM plan limits | 14 Aug 2026 credit change, throttling complaints |