From fd03e18fd020112a83b6257600d1b02baa2d07a8 Mon Sep 17 00:00:00 2001 From: "Joan Estrada Ramos (automation)" Date: Sat, 30 May 2026 13:17:00 +0000 Subject: [PATCH 1/3] Fix Cohere: replace non-existent command-r-plus-2025 with command-r-plus-08-2024 cohere.com/pricing lists no "2025" Command R+. Canonical current SKU is Command R+ 08-2024 at $2.50/$10 (the stored $3/$15 matched the older 04-2024 SKU). Migrates id, name, releaseDate, and pricing, and re-points benchmarks and use-cases. Adds dated research report. Co-Authored-By: Claude Opus 4.7 --- research/research_report_2026-05-30.md | 42 ++++++++++++++++++++++++++ src/data/benchmarks.ts | 2 +- src/data/models.ts | 6 ++-- src/data/pricing.ts | 8 ++--- src/data/use-cases.ts | 4 +-- 5 files changed, 52 insertions(+), 10 deletions(-) create mode 100644 research/research_report_2026-05-30.md diff --git a/research/research_report_2026-05-30.md b/research/research_report_2026-05-30.md new file mode 100644 index 0000000..b61f379 --- /dev/null +++ b/research/research_report_2026-05-30.md @@ -0,0 +1,42 @@ +# Daily AI vendor refresh — 2026-05-30 (escalation 4753e843) + +Locked providers: Anthropic, OpenAI, Google, Meta, Mistral, DeepSeek, xAI, Cohere. + +Escalation run 4753e843 flagged potential material updates across five vendors. Each candidate was re-verified against primary / authoritative official sources before any change. This report covers the **Cohere** correction made on this branch; the **Anthropic** (Opus 4.8) and **xAI** (grok-4 → grok-4.3) changes were folded into the existing open PRs (#7 and #8 respectively) to avoid duplicate PRs. + +## Diff & Freshness + +### Cohere — corrected (this branch) +- The dataset tracked a non-existent SKU `command-r-plus-2025` ($3 / $15). Cohere's pricing page lists **no "2025" Command R+**. The real, current SKUs are **Command R+ 08-2024 ($2.50 in / $10 out)** and the older **Command R+ 04-2024 ($3 in / $15 out)**. Source: https://cohere.com/pricing +- Decision: migrate the entry to the canonical latest **`command-r-plus-08-2024`** at **$2.50 / $10** (release 2024-08-30). Cross-references in benchmarks and use-cases updated to the new id. The old $3/$15 value matched the older 04-2024 SKU, confirming the stored id/price pairing was stale/inconsistent. + +### Anthropic — handled in PR #7 (data-refresh-2026-05-28) +- Added **Claude Opus 4.8** (`claude-opus-4-8`): 1M ctx, 128K max output, $5/$25, cache read $0.50, batch $2.50/$12.50. It is now Anthropic's flagship ("NextOpus"); Opus 4.7 is legacy. Opus 4.6/4.5 (also $5/$25) are legacy SKUs not tracked — not added to avoid dataset bloat. Sources: https://platform.claude.com/docs/en/docs/about-claude/models , .../pricing + +### xAI — handled in PR #8 (data-refresh-2026-05-29) +- `grok-4` was **retired 2026-05-15** and now redirects to **`grok-4.3`** at standard rates ($1.25 / $2.50, $0.20 cached, 1M ctx). The proposed $3/$15 change (from the x.ai/api marketing page) was rejected as a stale secondary source; the entry was migrated to the live `grok-4-3` id. Source: https://docs.x.ai/docs/models + +### Google — inconclusive earlier, now confirmed UNCHANGED +- The escalation noted absent paid-tier prices in a prior fetch. Re-fetch of https://ai.google.dev/gemini-api/docs/pricing shows paid-tier prices that **match the current dataset exactly**: Gemini 2.5 Pro $1.25 in / $10 out (≤200k), Gemini 2.5 Flash $0.30 in / $2.50 out. **No change.** + +### Meta — held back (no clean primary metadata) +- Official Meta (llama.com) remains robots-blocked. Azure AI Foundry's model catalog confirms Llama 4 Maverick exists as a natively-multimodal MoE but **does not publish numeric context window / max output** values. No reliable primary number found, so Llama 4 Maverick metadata is **left unchanged**. Source checked: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/models-featured + +## Models + +### Cohere +- `command-r-plus-08-2024` (Command R+ 08-2024): 128K context; $2.50 input / $10 output per 1M tokens. Source: https://cohere.com/pricing + +## Pricing + +### Cohere +- `command-r-plus-08-2024`: $2.50 input / $10.00 output per 1M tokens (list). Source: https://cohere.com/pricing + +## Benchmarks +- No new benchmark source integrated; existing Command R+ row re-pointed to `command-r-plus-08-2024` (scores carried over). + +## Use-cases +- Enterprise-RAG and regulated-EU use cases re-pointed from `command-r-plus-2025` to `command-r-plus-08-2024`. + +## Historic-usage +- No dataset changes made today. diff --git a/src/data/benchmarks.ts b/src/data/benchmarks.ts index 6edb3db..90a4644 100644 --- a/src/data/benchmarks.ts +++ b/src/data/benchmarks.ts @@ -181,7 +181,7 @@ export const benchmarks: Benchmark[] = [ source: "⚠ vendor-reported. Reasoning-tier sibling of V4 Flash; pending independent eval.", }, { - modelId: "command-r-plus-2025", + modelId: "command-r-plus-08-2024", reasoning: 76, coding: 70, math: 65, diff --git a/src/data/models.ts b/src/data/models.ts index 8f0a18b..aa54877 100644 --- a/src/data/models.ts +++ b/src/data/models.ts @@ -287,11 +287,11 @@ export const models: Model[] = [ "DeepSeek's reasoning-tier V4. Currently discounted 75% through 2026-05-31 — pricing will step up afterwards.", }, { - id: "command-r-plus-2025", - name: "Command R+ 2025", + id: "command-r-plus-08-2024", + name: "Command R+ 08-2024", provider: "Cohere", family: "Command R+", - releaseDate: "2025-06-12", + releaseDate: "2024-08-30", contextWindowTokens: 128_000, maxOutputTokens: 8_000, modalities: ["text", "tool-use"], diff --git a/src/data/pricing.ts b/src/data/pricing.ts index 8308683..aed2299 100644 --- a/src/data/pricing.ts +++ b/src/data/pricing.ts @@ -104,9 +104,9 @@ export const pricing: Pricing[] = [ notes: "75% promotional discount through 2026-05-31 15:59 UTC (per DeepSeek API docs).", }, { - modelId: "command-r-plus-2025", - inputPerMTokUsd: 3.0, - outputPerMTokUsd: 15.0, - notes: "Cohere Command R+ pricing (Cohere pricing page).", + modelId: "command-r-plus-08-2024", + inputPerMTokUsd: 2.5, + outputPerMTokUsd: 10.0, + notes: "Cohere Command R+ 08-2024 list pricing per cohere.com/pricing ($2.50 in / $10 out per 1M tokens).", }, ]; diff --git a/src/data/use-cases.ts b/src/data/use-cases.ts index 9738253..6f6312f 100644 --- a/src/data/use-cases.ts +++ b/src/data/use-cases.ts @@ -25,7 +25,7 @@ export const useCases: UseCase[] = [ id: "enterprise-rag", title: "Enterprise RAG with citations", category: "rag", - recommendedModels: ["command-r-plus-2025", "claude-sonnet-4-6", "gemini-2-5-pro"], + recommendedModels: ["command-r-plus-08-2024", "claude-sonnet-4-6", "gemini-2-5-pro"], why: "Cohere's Command R+ is purpose-built for grounded answers with citations. Gemini 2.5 Pro shines when the corpus is pasted into a single 1M+ token context. Sonnet is the safe middle.", costStrategy: @@ -65,7 +65,7 @@ export const useCases: UseCase[] = [ id: "regulated-eu-deployment", title: "Regulated EU / data-residency deployment", category: "support", - recommendedModels: ["mistral-large-3", "command-r-plus-2025"], + recommendedModels: ["mistral-large-3", "command-r-plus-08-2024"], why: "Mistral hosts in EU regions. Cohere offers private deployment options for regulated industries (finance, healthcare).", costStrategy: From 18540bfe6efd3775ec71020503b5fc87381c3537 Mon Sep 17 00:00:00 2001 From: Daily Vendor Refresh Bot Date: Mon, 1 Jun 2026 13:39:55 +0000 Subject: [PATCH 2/3] Document held-back items for 2026-06-01 escalation 4753e843 Records the conservative decisions to leave DeepSeek V4 Pro pricing unchanged (no explicit post-promo rate published; "1/4 of original" equals the current 75%-off values) and to hold back Grok Build 0.1. No dataset changes; Cohere correction on this branch is unaffected. Co-Authored-By: Claude Opus 4.7 --- research/research_report_2026-05-30.md | 15 +++++++++++++++ 1 file changed, 15 insertions(+) diff --git a/research/research_report_2026-05-30.md b/research/research_report_2026-05-30.md index b61f379..3c9fa39 100644 --- a/research/research_report_2026-05-30.md +++ b/research/research_report_2026-05-30.md @@ -40,3 +40,18 @@ Escalation run 4753e843 flagged potential material updates across five vendors. ## Historic-usage - No dataset changes made today. + +--- + +## Addendum — 2026-06-01 (escalation 4753e843 re-verification) + +Closing out the remaining escalation items. **No data changes** to this PR (Cohere correction stands as-is). + +### DeepSeek V4 Pro post-promo — held back (no explicit rate published) +- The DeepSeek pricing page (https://api-docs.deepseek.com/quick_start/pricing) states the 75% promotional discount for `deepseek-v4-pro` ends **2026-05-31 15:59 UTC** and that pricing "will be officially adjusted to **1/4 of the original price** after the promotion ends." +- The page does **not list explicit post-promo dollar amounts**. Critically, "75% discount" and "1/4 of the original price" are arithmetically the same multiplier (×0.25), so the snippet is self-contradictory about whether the rate actually changes. The current stored values ($0.435 in / $0.87 out / $0.003625 cache-hit) already equal 1/4 of the implied original. +- **Decision: dataset left unchanged.** Per the conservative source-backed policy, we do not infer a new post-promo number that the primary source does not explicitly state. No DeepSeek PR was opened (no existing PR touches DeepSeek, and the bar for a new PR — primary docs explicitly listing current rates — is not met). +- Follow-up: once the promo window has fully lapsed and DeepSeek publishes explicit standard rates, update `deepseek-v4-pro` pricing and refresh its model blurb (which still reads "discounted 75% through 2026-05-31"). Tracked as a held-back item, not actioned today. + +### Grok Build 0.1 — held back +- Documented in PR #8's report (research/research_report_2026-05-29.md): specialty xAI coding SKU, outside the locked one-flagship-per-provider strategy and missing schema-required metadata. Not added. From 1bf18a3881e29e8d0f9e90c902fc3e593591bd5b Mon Sep 17 00:00:00 2001 From: Joan Estrada Ramos Date: Wed, 3 Jun 2026 13:06:19 +0000 Subject: [PATCH 3/3] Correct Mistral Large 3 pricing to $0.5/$1.5 (2026-06-03 escalation 4753e843) Official Mistral pricing page now lists Mistral Large 3 at $0.5 input / $1.5 output per 1M tokens (was $2/$6). Updates mistral-large-3 entry and derives batch rate at 50% ($0.25). Maps cleanly to existing schema/id. Folded into existing PR #9 to avoid a duplicate daily-refresh PR; #6/#7/#8 left unmodified. Source: https://mistral.ai/pricing/ Co-Authored-By: Claude Opus 4.7 --- research/research_report_2026-05-30.md | 13 +++++++++++++ src/data/pricing.ts | 7 ++++--- 2 files changed, 17 insertions(+), 3 deletions(-) diff --git a/research/research_report_2026-05-30.md b/research/research_report_2026-05-30.md index 3c9fa39..7fb9970 100644 --- a/research/research_report_2026-05-30.md +++ b/research/research_report_2026-05-30.md @@ -55,3 +55,16 @@ Closing out the remaining escalation items. **No data changes** to this PR (Cohe ### Grok Build 0.1 — held back - Documented in PR #8's report (research/research_report_2026-05-29.md): specialty xAI coding SKU, outside the locked one-flagship-per-provider strategy and missing schema-required metadata. Not added. + +--- + +## Addendum — 2026-06-03 (escalation 4753e843 — Mistral Large 3 repricing) + +### Mistral — corrected (this branch) +- The dataset tracked `mistral-large-3` at **$2.00 in / $6.00 out** (batch $1.00). The official Mistral pricing page now lists **Mistral Large 3 at $0.5 input / $1.5 output per 1M tokens** ("Open-weight, general-purpose, flagship multimodal and multilingual model"). Source: https://mistral.ai/pricing/ (verified 2026-06-03). +- **Decision: updated** `mistral-large-3` to **$0.5 / $1.5**. The change maps cleanly to the existing `Pricing` schema and model id — no new id, no schema change required. +- **Batch discount:** Mistral's page states batch processing gets a general **50% discount** but does not enumerate a per-model batch line for Large 3. The schema has an optional `batchInputPerMTokUsd` field; the prior entry carried batch = 50% of input. To stay consistent and conservative, `batchInputPerMTokUsd` was updated to **$0.25** (50% of the new $0.5 input). This is a derived value from the stated general discount, not an explicit per-model figure from Mistral; noted here per the source-backed policy. +- **No new PR opened.** Per the daily-refresh duplicate-avoidance strategy, this correction was folded into the existing open PR #9 (branch `data-refresh-2026-05-30`) rather than opening a new daily-refresh PR. PRs #6/#7/#8 left unmodified. + +### Other Mistral SKUs on the pricing page — not added +- The page also lists Mistral Medium 3.5 ($1.5/$7.5), Small 4 ($0.1/$0.3), Devstral 2 ($0.4/$2), Codestral ($0.3/$0.9). These are outside the locked one-flagship-per-provider dataset strategy and were **not** added to avoid dataset bloat. diff --git a/src/data/pricing.ts b/src/data/pricing.ts index aed2299..b397f4b 100644 --- a/src/data/pricing.ts +++ b/src/data/pricing.ts @@ -85,9 +85,10 @@ export const pricing: Pricing[] = [ }, { modelId: "mistral-large-3", - inputPerMTokUsd: 2.0, - outputPerMTokUsd: 6.0, - batchInputPerMTokUsd: 1.0, + inputPerMTokUsd: 0.5, + outputPerMTokUsd: 1.5, + batchInputPerMTokUsd: 0.25, + notes: "Official Mistral pricing (mistral.ai/pricing, 2026-06-03): $0.5/$1.5 per 1M tokens; batch processing gets 50% discount.", }, { modelId: "deepseek-v4-flash",