Monthly Empirical Benchmark & Pricing Audit · Published 2026-10-28
State of Agent Infrastructure Pricing, Launches & Reliability (October 2026 Edition)
100% of 920 active pricing rows verified within <=9 days (target <=14 days); 58 historical unverified rows quarantined from price-sort rankings and re-checked on 2026-10-28; 10 frontier model launches and 6 active API/provider retirements tracked.
Active Prices ≤14d
100%
920 / 980 rows active
Price Reviews Logged
1182
96 corrections-log entries
Verified Outcomes
18 Trials
56 Countersign receipts
Error Rescue Cards
28 Traps
SSR at /rescue/[slug] + /api/rescue
1. Active Shutdowns & SKU Retirements (Breakwire Watch)
Subscribe via JSON Feed (/api/lifecycle/feed.json) →sku_retirementEffective: 2026-11-17
[Shutdown 2026-11-17] google-ai: gemini-2.5-pro-preview-05-06, gemini-2.5-flash-preview-05-20, gemini-2.0-flash-preview-image-generation, gemini-exp-1206
Google Gemini legacy preview model snapshots (gemini-2.5-pro-preview-*, gemini-2.5-flash-preview-*, gemini-exp-*) retire on November 17, 2026.
Verified Migration: Pin stable GA model slugs gemini-2.5-pro, gemini-2.5-flash, or the newly launched gemini-3.8-flash ($0.20/$0.80 per 1M tokens) via @google/genai.
sku_retirementEffective: 2027-01-06
[Shutdown 2027-01-06] openai: tts-1, tts-1-hd, tts-1-1106, tts-1-hd-1106
OpenAI legacy first-generation text-to-speech models (tts-1 and tts-1-hd) are scheduled for retirement on January 6, 2027 in favor of steerable instruction-capable audio models and dedicated low-latency TTS providers.
Verified Migration: Upgrade in-place to OpenAI gpt-4o-mini-tts, or switch to ElevenLabs (v4 Turbo TTS at $0.011/1k chars) or Deepgram Aura-2 for production voice agents.
sku_retirementEffective: 2026-10-31
[Shutdown 2026-10-31] smithery: @smithery/cli, @smithery/sdk, run.tools, server.smithery.ai
Smithery connection management services (hosted server URLs on run.tools and server.smithery.ai, Smithery API, Connect, Toolbox, accounts/API keys, CLI, and SDKs) shut down October 31, 2026. smithery.ai remains a non-proxied public directory.
Verified Migration: Replace run.tools / server.smithery.ai proxy URLs with direct remote Streamable-HTTP MCP server URLs discovered via Official MCP Registry (/.well-known/mcp/server.json) or host your own MCP server on Cloudflare Workers.
provider_retirementEffective: 2026-10-31
[Provider Retired 2026-10-31] smithery
On October 31, 2026, we're shutting down Smithery's connection management services. Smithery will remain as a public (non-proxied) MCP registry. Smithery says smithery.ai stays a public (non-proxied) MCP directory, 'now powered by Arcade.dev', with verified server pages that show direct MCP URLs. The page names Arcade.dev as its directory provider; Arcade is not a catalog provider here.
Verified Migration: Smithery says smithery.ai stays a public (non-proxied) MCP directory, 'now powered by Arcade.dev', with verified server pages that show direct MCP URLs. The page names Arcade.dev as its directory provider; Arcade is not a catalog provider here.
provider_retirementEffective: 2025-08-11
[Provider Retired 2025-08-11] bing-search
Bing Search APIs will be retired on August 11, 2025. Any existing instances of Bing Search APIs will be decommissioned completely, and the product will no longer be available to be used or new customer signup. Microsoft points customers to 'Grounding with Bing Search' in Azure AI Agents (not a catalog provider here).
Verified Migration: Microsoft points customers to 'Grounding with Bing Search' in Azure AI Agents (not a catalog provider here).
provider_retirementEffective: 2026-03-03
[Provider Retired 2026-03-03] helicone
Helicone will continue operating in maintenance mode—security updates, bug fixes, and new models will keep shipping. We'll work closely with every customer to support a smooth migration to another platform. No shutdown date or successor named. Mintlify's post (datePublished 2026-03-03, re-fetched 2026-10-21) says it will support migration 'to another platform'. Helicone's own letter (https://www.helicone.ai/blog/joining-mintlify) still returned HTTP 500 on 2026-10-21 and is not quoted.
Verified Migration: No shutdown date or successor named. Mintlify's post (datePublished 2026-03-03, re-fetched 2026-10-21) says it will support migration 'to another platform'. Helicone's own letter (https://www.helicone.ai/blog/joining-mintlify) still returned HTTP 500 on 2026-10-21 and is not quoted.
2. Frontier Model Launches & Verified Token Pricing (9)
| Launch Date | Provider | Model & Sourced Rate | Highlights & Replaced Models |
|---|---|---|---|
| 2026-10-24 | google-ai | [Model Launch] google-ai gemini-3.8-flash ($0.20 / 1M input tokens (text/image/video), $0.80 / 1M output tokens ($0.02 / 1M cached input)) | 1M-context multimodal Flash generation replacing retiring Gemini 2.5 preview snapshots (retiring 2026-11-17). Pricing: $0.20 / 1M input tokens (text/image/video), $0.80 / 1M output tokens ($0.02 / 1M cached input). Replaces: gemini-2.5-flash-preview-05-20, gemini-1.5-flash. |
| 2026-10-24 | google-ai | [Model Launch] google-ai gemini-3.8-live ($2.00 / 1M audio input tokens, $8.00 / 1M audio output tokens; Gemini 3.8 Flash TTS at $9.00 / 1M audio output tokens) | Native bidirectional audio/video streaming and Flash TTS. Pricing: $2.00 / 1M audio input tokens, $8.00 / 1M audio output tokens; Gemini 3.8 Flash TTS at $9.00 / 1M audio output tokens. Replaces: gemini-2.0-flash-live. |
| 2026-10-24 | openai | [Model Launch] openai gpt-6.1-sol ($2.50 / 1M input tokens, $12.00 / 1M output tokens ($0.25 / 1M cached input)) | OpenAI flagship frontier reasoning + tool-use model with 90% cached input discount. Pricing: $2.50 / 1M input tokens, $12.00 / 1M output tokens ($0.25 / 1M cached input). Replaces: gpt-4o, gpt-4-turbo. |
| 2026-10-24 | openai | [Model Launch] openai gpt-6-astra ($0.60 / 1M input tokens, $3.00 / 1M output tokens ($0.06 / 1M cached input)) | Mid-tier high-throughput workhorse for multi-step agent loops. Pricing: $0.60 / 1M input tokens, $3.00 / 1M output tokens ($0.06 / 1M cached input). Replaces: gpt-4o-mini. |
| 2026-10-24 | openai | [Model Launch] openai gpt-6-luna ($0.12 / 1M input tokens, $0.50 / 1M output tokens ($0.012 / 1M cached input)) | Ultra-low-cost classifier, router, and extraction model. Pricing: $0.12 / 1M input tokens, $0.50 / 1M output tokens ($0.012 / 1M cached input). Replaces: gpt-3.5-turbo. |
| 2026-10-20 | minimax | [Model Launch] minimax minimax-m3.1-flash ($0.10 / 1M input tokens ($0.015 / 1M cache read), $0.45 / 1M output tokens) | Ultra-cheap long-context agent model with $0.015/1M cache reads. Pricing: $0.10 / 1M input tokens ($0.015 / 1M cache read), $0.45 / 1M output tokens. Replaces: minimax-text-01. |
| 2026-10-18 | deepseek | [Model Launch] deepseek deepseek-v4-flash ($0.16 / 1M input tokens ($0.02 / 1M cache hit), $0.68 / 1M output tokens) | High-speed reasoning + coding open-weights API with automatic context caching. Pricing: $0.16 / 1M input tokens ($0.02 / 1M cache hit), $0.68 / 1M output tokens. Replaces: deepseek-chat. |
| 2026-10-15 | anthropic | [Model Launch] anthropic claude-opus-4-5 ($5.00 / 1M input tokens, $25.00 / 1M output tokens (67% cheaper than legacy Claude 3 Opus)) | 67% price cut vs legacy Opus ($5/$25 vs $15/$75) with frontier deep reasoning. Pricing: $5.00 / 1M input tokens, $25.00 / 1M output tokens (67% cheaper than legacy Claude 3 Opus). Replaces: claude-3-opus-20240229. |
| 2026-09-29 | anthropic | [Model Launch] anthropic claude-sonnet-4-5 ($3.00 / 1M input tokens, $15.00 / 1M output tokens ($0.30 / 1M cache read)) | Depot Pick SKU for coding-agent and autonomous tool loops; replaces deprecated claude-3-5-sonnet-* snapshots. Pricing: $3.00 / 1M input tokens, $15.00 / 1M output tokens ($0.30 / 1M cache read). Replaces: claude-3-5-sonnet-20241022, claude-3-5-sonnet-20240620. |
3. Top Runtime Failure Rescue Signatures (12 of 28)
View all 28 rescue cards →deprecated_model
Deprecated Anthropic Claude 3.5 / 3.x Snapshot (404 model_not_found)
Fix: claude-sonnet-4-5
deprecated_model
Deprecated / Overpriced OpenAI GPT-4 & GPT-3.5 Snapshots
Fix: gpt-4.1
deprecated_model
Legacy OpenAI text-embedding-ada-002 (5x Overpriced Embedding SKU)
Fix: text-embedding-3-small
wrong_package
Deprecated @google/generative-ai SDK & Gemini 1.5 / 1.0 Models
Fix: @google/genai (gemini-2.5-flash / gemini-2.5-pro)
wrong_package
Wrong npm Package Trap: 'r2' (Archived 2017 HTTP Client, NOT Cloudflare R2)
Fix: @aws-sdk/client-s3 (r2-standard-storage)
wrong_package
Deprecated Package Trap: 'pinecone-client' vs '@pinecone-database/pinecone'
Fix: @pinecone-database/pinecone
wrong_package
Deprecated Package Trap: 'mailgun-js' (Unmaintained SDK)
Fix: resend
rate_limit_429
HTTP 429 Too Many Requests / Token Quota Exhaustion
Fix: prompt-caching + openrouter / google-ai fallback
provider_outage
HTTP 529 Overloaded / Upstream Model Outage
Fix: gemini-2.5-pro (google-ai) or openrouter fallback
pricing_trap
HTTP 402 Payment Required / Prepaid Credit Exhaustion
Fix: gemini-2.5-flash (free tier) or groq free tier
waf_bot_block
WAF / Cloudflare 1020 / 403 Forbidden / robots.txt Disallow on Web Fetch
Fix: firecrawl (web-scrape) or browserbase (cloud-browser)
platform_limit
SMTP Connection Timeout on Port 25 / 465 / 587
Fix: resend HTTP API (port 443)
4. Recent Double-Curl Pricing Audit Log (October 2026)
| Date | Entry ID | Provider | Audit Finding |
|---|---|---|---|
| 2026-10-28 | fix-0.14.0-bedrock-claude-3-5-sonnet-still-unverified | Re-checked https://aws.amazon.com/bedrock/pricing/ on 2026-10-28 (~20:14 ET, fetched twice, nonce-normalized text sha256 a94d9435...f949): Claude 3.5 Sonnet and Claude 3.5 Sonnet v2 (6 rows) remain unrendered JS priceOf placeholders in static HTML. All 6 rows remain sourceStatus=unverified with sourceStatusCheckedOn=2026-10-28; 0 amounts, units, sourceUrls, asOf, scores or picks changed. | |
| 2026-10-28 | fix-0.14.0-minimax-m2x-still-unverified | Re-checked https://platform.minimax.io/docs/pricing/overview (308 target of pricing-paygo) on 2026-10-28 (~20:14 ET, fetched twice, byte-identical sha256 c9f0db96...852d): MiniMax-M2, MiniMax-M2.1, MiniMax-M2.5 and MiniMax-M2.7 (28 rows) are still absent from the LLM pricing tables. All 28 rows remain sourceStatus=unverified with sourceStatusCheckedOn=2026-10-28; 0 amounts, units, sourceUrls, asOf, scores or picks changed. | |
| 2026-10-28 | fix-0.14.0-friendli-minimax-m25-still-unverified | Re-checked https://friendli.ai/pricing on 2026-10-28 (~20:14 ET, fetched twice, byte-identical sha256 f74455be...453a): MiniMax-M2.5 (input/output/cached) is still not on the Pay per token table. All 3 rows remain sourceStatus=unverified with sourceStatusCheckedOn=2026-10-28; 0 amounts, units, sourceUrls, asOf, scores or picks changed. | |
| 2026-10-28 | fix-0.13.4-elevenlabs-entity-detection-still-unverified | ||
| 2026-10-28 | fix-0.13.4-openai-gpt-6-luna-long-still-unverified | ||
| 2026-10-28 | fix-0.13.4-sambanova-minimax-m3-still-unverified | ||
| 2026-10-28 | fix-0.13.4-chutes-still-unverified | ||
| 2026-10-28 | fix-0.12.164-e2b-reconfirm | ||
| 2026-10-28 | fix-0.12.163-composio-reconfirm | ||
| 2026-10-28 | fix-0.12.162-google-ai-reconfirm |