fix(anthropic): align Claude 4.x fallback pricing & context limits to current rates by KennethWKZ · Pull Request #1485 · headroomlabs-ai/headroom

KennethWKZ · 2026-06-27T02:47:37Z

Summary

Aligns the Anthropic fallback pricing and context-limit tables in headroom/providers/anthropic.py with the current anthropic.com/pricing rates (verified 2026-06-27). Several Claude 4.x entries were carrying stale or incorrect rates, and the newer Sonnet 4.5 / 4.6 models had no fallback metadata at all.

Changes

All values $ / MTok; cached_input = prompt-cache read = 0.1× input.

Tier	Model	Before	After	Context
Opus	`claude-opus-4-8`	$15 / $75 / $1.50	$5 / $25 / $0.50	1M
Opus	`claude-opus-4-7`	$15 / $75 / $1.50	$5 / $25 / $0.50	1M
Opus	`claude-opus-4-6`	$15 / $75 / $1.50	$5 / $25 / $0.50	1M
Opus	`claude-opus-4-5-20251101`	$15 / $75 / $1.50	$5 / $25 / $0.50	200K
Sonnet	`claude-sonnet-4-6`	— (new)	$3 / $15 / $0.30	1M
Sonnet	`claude-sonnet-4-5`	— (new)	$3 / $15 / $0.30	200K
Haiku	`claude-haiku-4-5-20251001`	$0.80 / $4 / $0.08 (3.5 rates)	$1 / $5 / $0.10	200K

claude-sonnet-4-20250514 and all Claude 3.x / 3.5.x entries were already correct — left unchanged.

The Sonnet 4.6 entry is functional, not cosmetic: it ships in the long-context pricing tier (1M window), and without an explicit entry the sonnet pattern default would report 200K.

Tests

Replaced the original fragile parity-with-Opus-4.7 assertion with parametrized absolute-value tier checks:

test_current_opus_tier_pricing — Opus 4.5–4.8
test_current_sonnet_tier_pricing — Sonnet 4 / 4.5 / 4.6
test_current_haiku_tier_pricing — Haiku 4.5
test_get_context_limit_claude_sonnet_4_6 — pins the 1M context window

Full suite: 24 passed, ruff clean.

Notes

_PATTERN_DEFAULTS (the opus / sonnet / haiku regex fallbacks for unknown models) are intentionally left unchanged — they are conservative heuristics, not model-specific entries. Happy to retune in a follow-up.
Source of truth: https://docs.anthropic.com/en/docs/about-claude/pricing

Register the newest Opus tier (claude-opus-4-8) in ANTHROPIC_CONTEXT_LIMITS (1M) and ANTHROPIC_PRICING (Opus tier: $15/$75/$1.50 per Mtok), matching the claude-opus-4-7 entry it sits beside. Without this, the proxy cannot size the context window or estimate cost for the model. Add context-limit and pricing-parity tests alongside the existing 4-7 ones.

github-actions · 2026-06-27T02:47:51Z

PR governance

This PR does not yet satisfy the required template fields:

Missing required section Real Behavior Proof.
Missing required section Review Readiness.
Check at least one verification item in Testing.
Paste real command output or artifact links in Testing → Test Output.
Check I have performed a self-review before requesting human review.
Check This PR is ready for human review or convert the PR back to draft.

Please update the PR body, or move the PR back to draft while it is still in progress.

…s, aws login deps Layer the team's tooling on top of mhaitana's feat/bedrock-application-inference-profile-arn-support (ARN converse routing, named-profile wiring to the LiteLLM completion calls, au. region prefix, regression tests, and the Claude Code + Bedrock guide that resolves the review blockers): - headroom/providers/anthropic.py: register claude-opus-4-8 context limit (1M) + Opus-tier pricing (from headroomlabs-ai#1485). - pyproject.toml: bedrock extra = boto3>=1.41.0 + botocore[crt]>=1.41.0. aws login (IAM Identity Provider / console-login, DPoP) requires boto3 1.41+ with the AWS Common Runtime (CRT) per AWS docs; include bedrock in [all]. - docker-compose.bedrock.yml: Bedrock overlay; passes --bedrock-profile explicitly so the named profile reaches LiteLLM completions (not just startup discovery); mounts host ~/.aws read-only for the login token cache. - docker-compose.provider.yml: generic Anthropic-compatible upstream overlay (GLM/z.ai, DeepSeek). - docs/bedrock-cli-setup.md: CLI (non-Docker) Bedrock guide; documents CLAUDE_CODE_USE_BEDROCK=0 to prevent Claude Code bypassing the proxy. - .env.example: documented overlay vars.

JerrettDavis

Thanks for adding the new model metadata. The context limit matches the current Anthropic docs, but the pricing is stale for Opus 4.8: Anthropic's current pricing page lists Claude Opus 4.8 at $5 / MTok input and $25 / MTok output, with cached input at the corresponding cache-write/read rates, not the older $15/$75 Opus tier used here. The PR body and parity test also say it should match 4.7, which is exactly the part that needs updating now.

Please update ANTHROPIC_PRICING["claude-opus-4-8"] and the test expectation to the current Opus 4.8 values. Reference: https://docs.anthropic.com/en/docs/about-claude/pricing

Opus 4.8, 4.7, and 4.6 were stored at the deprecated Opus tier ($15 input / $75 output / $1.50 cache-read), but all three have shipped at the current Opus tier since 4.5: $5/MTok input, $25/MTok output, and a $0.50/MTok cache read (0.1x input). Verified against the live anthropic.com/pricing table. Replaces the fragile parity-with-4.7 assertion with explicit absolute-value checks across 4.6-4.8 so the rate is pinned rather than silently inherited. Refs headroomlabs-ai#1485

KennethWKZ · 2026-06-27T05:47:27Z

This now aligns the entire Claude 4.x line in the fallback pricing / context-limit tables to current rates:

Opus tier — $5 / $25 / $0.50 cache-read (was the deprecated $15 / $75 / $1.50 Opus 4.1 tier):

claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101

Sonnet tier — $3 / $15 / $0.30 cache-read:

claude-sonnet-4-6 (new — and functional: ships with the 1M context window; the sonnet pattern default would otherwise report 200K)
claude-sonnet-4-5 (new, 200K)
claude-sonnet-4-20250514 (unchanged, already correct)

Haiku tier — $1 / $5 / $0.10 cache-read:

claude-haiku-4-5-20251001 (was carrying Haiku 3.5 rates $0.80 / $4 / $0.08)

Context windows: added Sonnet 4.6 = 1M (long-context tier); Opus 4.5–4.8 already correct.

Tests: replaced the original fragile parity-with-4.7 assertion with parametrized absolute-value tier checks (Opus 4.5–4.8, Sonnet 4–4.6, Haiku 4.5) plus a Sonnet 4.6 1M-context test.

Left intentionally unchanged: _PATTERN_DEFAULTS (the opus/sonnet/haiku regex fallbacks). They're conservative heuristic estimates for unknown models, not model-specific entries — happy to retune them too if you want unknown-model estimates to track current tiers.

Happy to split into focused PRs if you'd rather keep the review surface tighter.

Opus 4.8, 4.7, and 4.6 were stored at the deprecated Opus tier ($15 input / $75 output / $1.50 cache-read), but all three have shipped at the current Opus tier since 4.5: $5/MTok input, $25/MTok output, and a $0.50/MTok cache read (0.1x input). Verified against the live anthropic.com/pricing table. Replaces the fragile parity-with-4.7 assertion with explicit absolute-value checks across 4.6-4.8 so the rate is pinned rather than silently inherited. Refs headroomlabs-ai#1485

Extend the fallback model tables to cover the newer Sonnet releases and correct the last stale Opus entry, all per the current anthropic.com/pricing page (verified 2026-06-27): - claude-sonnet-4-6: $3/$15/$0.30, 1M context window (ships in the long-context pricing tier — the sonnet pattern default would otherwise report 200K) - claude-sonnet-4-5: $3/$15/$0.30, 200K context - claude-opus-4-5-20251101: corrected from the deprecated $15/$75 Opus tier to the current $5/$25/$0.50 tier shared by 4.5-4.8 Adds Sonnet context-limit and tier-pricing test coverage; extends the Opus tier test to include 4.5. Refs headroomlabs-ai#1485

claude-haiku-4-5-20251001 was carrying Haiku 3.5 rates ($0.80/$4/$0.08); the current anthropic.com/pricing page lists Haiku 4.5 at $1/MTok input, $5/MTok output, cache read $0.10 (0.1x input). Adds a Haiku tier test. Refs headroomlabs-ai#1485

KennethWKZ · 2026-06-27T05:59:07Z

Consolidated update — supersedes my earlier piecemeal comments. All rates verified against the live anthropic.com/pricing table on 2026-06-27.

This now aligns the entire Claude 4.x line in the fallback pricing / context-limit tables to current rates:

Opus tier — $5 / $25 / $0.50 cache-read (was the deprecated $15 / $75 / $1.50 Opus 4.1 tier):

claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101

Sonnet tier — $3 / $15 / $0.30 cache-read:

claude-sonnet-4-6 (new — and functional: ships with the 1M context window; the sonnet pattern default would otherwise report 200K)
claude-sonnet-4-5 (new, 200K)
claude-sonnet-4-20250514 (unchanged, already correct)

Haiku tier — $1 / $5 / $0.10 cache-read:

claude-haiku-4-5-20251001 (was carrying Haiku 3.5 rates $0.80 / $4 / $0.08)

Context windows: added Sonnet 4.6 = 1M (long-context tier); Opus 4.5–4.8 already correct.

Tests: replaced the original fragile parity-with-4.7 assertion with parametrized absolute-value tier checks (Opus 4.5–4.8, Sonnet 4–4.6, Haiku 4.5) plus a Sonnet 4.6 1M-context test.

Left intentionally unchanged: _PATTERN_DEFAULTS (the opus/sonnet/haiku regex fallbacks). They are conservative heuristic estimates for unknown models, not model-specific entries — happy to retune them too if you want unknown-model estimates to track current tiers.

Happy to split into focused PRs if you would rather keep the review surface tighter.

KennethWKZ

done

## Description `pip install headroom-ai[bedrock]` cannot serve users who authenticate with `aws login` (IAM Identity Provider / console-login, DPoP). Resolving those credentials requires the AWS Common Runtime (CRT); without `awscrt`, botocore raises `MissingDependencyException`. The AWS docs state the requirement as: **"Boto3 version 1.41.0 or later with AWS Common Runtime (CRT)"** — i.e. both a modern boto3 floor and CRT (installed via the `[crt]` extra). ## Type of Change - [x] Bug fix (non-breaking) ## Changes Made - `pyproject.toml` `bedrock` extra: bump `boto3>=1.28.0` → `boto3>=1.41.0`, add `botocore[crt]>=1.41.0` (installs `awscrt`). - `uv.lock`: regenerated — adds `awscrt`, resolves `boto3` to 1.42.x. No code changes — the bedrock backend already passes `aws_profile_name` through to the LiteLLM calls (via #1456); this just makes the installed dependencies actually able to resolve `aws login` credentials. ## Impact - **`aws login` (IAM Identity Provider / DPoP):** now works — awscrt present. - **`aws sso login` (classic Identity Center):** unaffected (already worked). - **static keys (`~/.aws/credentials`):** unaffected. - Bumping the boto3 floor only affects the optional `[bedrock]` extra; bedrock users benefit from a current boto3 regardless. ## Testing Dependency-only change. `uv lock` resolves cleanly (257 packages, awscrt 0.29.2, boto3 1.42.38). No runtime code path altered, so existing bedrock tests are unaffected. ## Checklist - [x] Self-review performed - [x] No new warnings - [x] Linting passes ## Additional Notes Focused on the dependency gap only. ARN routing / named-profile wiring / docs are handled in #1456; pricing in #1485.

JerrettDavis

Requesting changes for the stale fallback-model tests. The pricing table changes match Anthropic's current model docs/pricing direction for Opus 4.x, Sonnet 4.x, and Haiku 4.5, but CI is still red because tests/test_provider_model_fallback.py expects the old Opus fallback rate:

TestAnthropicModelFallback.test_pricing_for_known_models still asserts $15/$75.
TestAnthropicModelFallback.test_cost_estimation_for_new_models still expects the corresponding $22.5 estimate; the updated rates correctly produce $7.5 for that fixture.

Please update those fallback-model tests to pin the new values too, so the provider-specific tests and the generic fallback tests agree.

Opus 4.8, 4.7, and 4.6 were stored at the deprecated Opus tier ($15 input / $75 output / $1.50 cache-read), but all three have shipped at the current Opus tier since 4.5: $5/MTok input, $25/MTok output, and a $0.50/MTok cache read (0.1x input). Verified against the live anthropic.com/pricing table. Replaces the fragile parity-with-4.7 assertion with explicit absolute-value checks across 4.6-4.8 so the rate is pinned rather than silently inherited. Refs headroomlabs-ai#1485

Extend the fallback model tables to cover the newer Sonnet releases and correct the last stale Opus entry, all per the current anthropic.com/pricing page (verified 2026-06-27): - claude-sonnet-4-6: $3/$15/$0.30, 1M context window (ships in the long-context pricing tier — the sonnet pattern default would otherwise report 200K) - claude-sonnet-4-5: $3/$15/$0.30, 200K context - claude-opus-4-5-20251101: corrected from the deprecated $15/$75 Opus tier to the current $5/$25/$0.50 tier shared by 4.5-4.8 Adds Sonnet context-limit and tier-pricing test coverage; extends the Opus tier test to include 4.5. Refs headroomlabs-ai#1485

claude-haiku-4-5-20251001 was carrying Haiku 3.5 rates ($0.80/$4/$0.08); the current anthropic.com/pricing page lists Haiku 4.5 at $1/MTok input, $5/MTok output, cache read $0.10 (0.1x input). Adds a Haiku tier test. Refs headroomlabs-ai#1485

Opus 4.8, 4.7, and 4.6 were stored at the deprecated Opus tier ($15 input / $75 output / $1.50 cache-read), but all three have shipped at the current Opus tier since 4.5: $5/MTok input, $25/MTok output, and a $0.50/MTok cache read (0.1x input). Verified against the live anthropic.com/pricing table. Replaces the fragile parity-with-4.7 assertion with explicit absolute-value checks across 4.6-4.8 so the rate is pinned rather than silently inherited. Refs headroomlabs-ai#1485

Extend the fallback model tables to cover the newer Sonnet releases and correct the last stale Opus entry, all per the current anthropic.com/pricing page (verified 2026-06-27): - claude-sonnet-4-6: $3/$15/$0.30, 1M context window (ships in the long-context pricing tier — the sonnet pattern default would otherwise report 200K) - claude-sonnet-4-5: $3/$15/$0.30, 200K context - claude-opus-4-5-20251101: corrected from the deprecated $15/$75 Opus tier to the current $5/$25/$0.50 tier shared by 4.5-4.8 Adds Sonnet context-limit and tier-pricing test coverage; extends the Opus tier test to include 4.5. Refs headroomlabs-ai#1485

claude-haiku-4-5-20251001 was carrying Haiku 3.5 rates ($0.80/$4/$0.08); the current anthropic.com/pricing page lists Haiku 4.5 at $1/MTok input, $5/MTok output, cache read $0.10 (0.1x input). Adds a Haiku tier test. Refs headroomlabs-ai#1485

KennethWKZ · 2026-07-03T16:22:50Z

Superseded by #1767 — rebased onto latest main, squashed to one commit, extended with Claude 5 family (Fable 5, Opus 4.8, Sonnet 5) and the requested fallback-test alignment.

…1767) ## Summary Adds Claude 5 generation metadata and aligns the Anthropic fallback pricing / context-limit tables in `headroom/providers/anthropic.py` with current [Anthropic pricing](https://platform.claude.com/docs/en/about-claude/pricing) (verified 2026-07-04). Several Claude 4.x entries carried stale rates, the new Claude 5 models (Fable 5, Opus 4.8, Sonnet 5) had no metadata, and the generic fallback tests disagreed with the provider tests. Supersedes #1485 (rebased onto latest `main`, squashed to one commit, extended with Sonnet 5 / Fable 5 and the requested fallback-test fixes). ## Changes All values `$ / MTok`; `cached_input` = prompt-cache read = 0.1× input. | Tier | Model | Before | After | Context | |---|---|---|---|---| | Fable | `claude-fable-5` | — *(new)* | $10 / $50 / $1.00 | **1M** | | Opus | `claude-opus-4-8` | — *(new)* | $5 / $25 / $0.50 | **1M** | | Opus | `claude-opus-4-7` | $15 / $75 / $1.50 | $5 / $25 / $0.50 | 1M | | Opus | `claude-opus-4-6` | $15 / $75 / $1.50 | $5 / $25 / $0.50 | 1M | | Opus | `claude-opus-4-5-20251101` | $15 / $75 / $1.50 | $5 / $25 / $0.50 | 200K | | Sonnet | `claude-sonnet-5` | — *(new)* | $3 / $15 / $0.30 | **1M** | | Sonnet | `claude-sonnet-4-6` | — *(new)* | $3 / $15 / $0.30 | **1M** | | Sonnet | `claude-sonnet-4-5` | — *(new)* | $3 / $15 / $0.30 | 200K | | Haiku | `claude-haiku-4-5-20251001` | $0.80 / $4 / $0.08 *(3.5 rates)* | $1 / $5 / $0.10 | 200K | `claude-sonnet-4-20250514` and all Claude 3.x / 3.5.x entries were already correct — left unchanged. The **1M-context** entries (Fable 5, Opus 4.8, Sonnet 5, Sonnet 4.6) are functional, not cosmetic: they ship in the long-context tier, and without explicit entries the `sonnet` / `opus` pattern defaults would report 200K. Sonnet 5 is pinned to the **standard** Sonnet tier ($3 / $15 / $0.30); Anthropic's introductory rate ($2 / $10 through Aug 31 2026) is intentionally not encoded to avoid a time-dependent fixture. ## Fallback-model tests (addresses review on #1485) `_PATTERN_DEFAULTS["opus"]` is aligned to the current Opus tier ($5 / $25 / $0.50) so the generic fallback suite and the provider-specific suite agree: - `test_pricing_for_known_models` — Opus 4.5 pins $5 / $25 / $0.50 - `test_pattern_based_inference_opus` — unknown-opus fallback now $5 / $25 - `test_cost_estimation_for_new_models` — fixture estimate corrected $22.5 → $7.5 - `test_pattern_based_inference_sonnet` — retargeted to `claude-sonnet-6-*` (the old `claude-sonnet-5-*` probe now prefix-matches the real `claude-sonnet-5` key) New provider coverage: - `test_get_context_limit_claude_5_family` — Fable 5 / Opus 4.8 / Sonnet 5 all 1M - `test_pricing_claude_5_family` — exact rate table for the 3 new models Full suite: **48 passed**. ## Source https://platform.claude.com/docs/en/about-claude/pricing

github-actions Bot added the status: needs author action Pull request body or readiness checklist still needs author updates label Jun 27, 2026

KennethWKZ mentioned this pull request Jun 27, 2026

feat(bedrock): support inference-profile ARNs and forward anthropic-beta (1M context) #1452

Closed

14 tasks

KennethWKZ mentioned this pull request Jun 27, 2026

fix(bedrock): add boto3 1.41 + CRT for aws login credentials #1486

Merged

4 tasks

JerrettDavis requested changes Jun 27, 2026

View reviewed changes

github-actions Bot added the status: ci failing Required or reported CI checks are failing label Jun 27, 2026

KennethWKZ changed the title ~~feat(anthropic): add claude-opus-4-8 context limit and pricing~~ fix(anthropic): align Claude 4.x fallback pricing & context limits to current rates Jun 27, 2026

github-actions Bot removed the status: ci failing Required or reported CI checks are failing label Jun 27, 2026

KennethWKZ commented Jun 27, 2026

View reviewed changes

KennethWKZ requested a review from JerrettDavis June 27, 2026 08:23

github-actions Bot added the status: ci failing Required or reported CI checks are failing label Jun 29, 2026

JerrettDavis requested changes Jun 29, 2026

View reviewed changes

KennethWKZ mentioned this pull request Jul 3, 2026

feat(anthropic): add Claude 5 family pricing & align current rates #1767

Merged

KennethWKZ closed this Jul 3, 2026

KennethWKZ deleted the feat/anthropic-opus-4-8-pricing branch July 3, 2026 16:28

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(anthropic): align Claude 4.x fallback pricing & context limits to current rates#1485

fix(anthropic): align Claude 4.x fallback pricing & context limits to current rates#1485
KennethWKZ wants to merge 4 commits into
headroomlabs-ai:mainfrom
KennethWKZ:feat/anthropic-opus-4-8-pricing

KennethWKZ commented Jun 27, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Jun 27, 2026

Uh oh!

JerrettDavis left a comment

Uh oh!

KennethWKZ commented Jun 27, 2026 •

edited

Loading

Uh oh!

KennethWKZ commented Jun 27, 2026

Uh oh!

KennethWKZ left a comment

Uh oh!

JerrettDavis left a comment

Uh oh!

KennethWKZ commented Jul 3, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

KennethWKZ commented Jun 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Changes

Tests

Notes

Uh oh!

github-actions Bot commented Jun 27, 2026

PR governance

Uh oh!

JerrettDavis left a comment

Choose a reason for hiding this comment

Uh oh!

KennethWKZ commented Jun 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

KennethWKZ commented Jun 27, 2026

Uh oh!

KennethWKZ left a comment

Choose a reason for hiding this comment

Uh oh!

JerrettDavis left a comment

Choose a reason for hiding this comment

Uh oh!

KennethWKZ commented Jul 3, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

KennethWKZ commented Jun 27, 2026 •

edited

Loading

KennethWKZ commented Jun 27, 2026 •

edited

Loading