Compare Claude Sonnet 4.6 and Claude Sonnet 5: Enterprise Migration and Quality Gains
Claude Sonnet 4.6 vs Claude Sonnet 5: which should I use?
Claude Sonnet 4.6 and Sonnet 5 share identical base pricing ($3.00/M input, $15.00/M output, $3.75 write / $0.30 read cache). Sonnet 5 provides generational improvements in agentic tool orchestration, coding benchmarks, and multi-turn stability. Verified 2026-09-07.
Verified 2026-09-07
Where can you find price, speed, and task evidence for Claude Sonnet 4.6 and Claude Sonnet 5?
Which tasks fit Claude Sonnet 4.6 and Claude Sonnet 5?
Best-fit task signals from the published model strengths; this is not a substitute for a controlled benchmark.
| Fact | Claude Sonnet 4.6 | Claude Sonnet 5 |
|---|---|---|
| Best fit | Enterprise workloads that need Opus-adjacent quality at Sonnet pricing. | Production workloads that need near-Opus quality with better speed and cost. |
| Reasoning mode | Available | Available |
What does a Coding Agent workload cost with Claude Sonnet 4.6 and Claude Sonnet 5?
Modeled for Coding Agent: 20,000 input + 2,000 output tokens per task, 1 turn. This is a workload model, not a provider quote.
| Fact | Claude Sonnet 4.6 | Claude Sonnet 5 |
|---|---|---|
| Coding Agent / task | $0.09 (modeled) | $0.06 (modeled) (winner) |
| Input / output rate | $3.00 / $15.00 per M | $2.00 / $10.00 per M |
How fast are Claude Sonnet 4.6 and Claude Sonnet 5?
Only non-estimated benchmark results are shown as measured.
| Fact | Claude Sonnet 4.6 | Claude Sonnet 5 |
|---|---|---|
| Measured throughput | 76 tokens/s | Unavailable |
| Time to first token | 360 ms | Unavailable |
How compatible are Claude Sonnet 4.6 and Claude Sonnet 5 with APIs?
Provider-level wire-format and SDK facts; model-specific parameter differences may still apply.
| Fact | Claude Sonnet 4.6 | Claude Sonnet 5 |
|---|---|---|
| OpenAI SDK | Not drop-in | Not drop-in |
| Request shape | Messages API — system prompt is a top-level field, not a message role, and max_tokens is required on every call. | Messages API — system prompt is a top-level field, not a message role, and max_tokens is required on every call. |
| Streaming | SSE (Anthropic content-block events) | SSE (Anthropic content-block events) |
What are the privacy and retention policies for Claude Sonnet 4.6 and Claude Sonnet 5?
Provider policy and residency facts from the verified provider registry. A missing retention policy is not treated as a guarantee.
| Fact | Claude Sonnet 4.6 | Claude Sonnet 5 |
|---|---|---|
| Provider says API data trains models | No | No |
| Published retention period | Unavailable | Unavailable |
| Data residency | Unavailable | Unavailable |
How much effort does it take to migrate between Claude Sonnet 4.6 and Claude Sonnet 5?
Direction is from each displayed model to the other. Effort is derived from provider compatibility and published parameter maps.
| Fact | Claude Sonnet 4.6 | Claude Sonnet 5 |
|---|---|---|
| Move Claude Sonnet 4.6 → Claude Sonnet 5 | drop-in; 0 breaking parameter differences | Target: Claude Sonnet 5 |
| Move Claude Sonnet 5 → Claude Sonnet 4.6 | Source: Claude Sonnet 5 | drop-in; 0 breaking parameter differences |
| Why | Same provider; only the model identifier changes. | Same provider; only the model identifier changes. |
Exact-pair decision and evidence guide. Verified 2026-09-01. These three sections are computed from dated scenarios; assumptions and unavailable joins are not observations.
Exact pair boundary: requested models are claude-sonnet-4-6 and claude-sonnet-5. A broad provider, pricing, task, or rate-limit verdict is out of scope.
Sonnet generation request-and-response delta compiler
Frozen fixture board. Formula / decision rule: delta = exact old/new ID + parameter + response field + support state; alias cannot mask unknown change Boundary: The compiler owns this pair’s upgrade decision; standalone contracts and migration tutorials remain elsewhere.
| Frozen fixture | Joined fields and evidence | Output | State |
|---|---|---|---|
| ordinary text | provider=Anthropic; host=api.anthropic.com; old=claude-sonnet-4-6; new=claude-sonnet-5; setting=text; effort=default; response=content blocks; evidence=2026-09-01 | supported/changed state=Unresolved until both response fields join | UNAVAILABLE — response delta |
| adaptive reasoning | provider=Anthropic; host=api.anthropic.com; old=claude-sonnet-4-6; new=claude-sonnet-5; setting=thinking=adaptive; budget=declared; output=answer+thinking; evidence=2026-09-01 | client edit=required if parameter or response field changed | CONDITIONAL — probe control/response |
| strict schema | provider=Anthropic; host=api.anthropic.com; old=claude-sonnet-4-6; new=claude-sonnet-5; setting=schema=strict; hash=declared; parser=versioned; evidence=2026-09-01 | schema parity=Unavailable; invalidate only after exact probe | UNAVAILABLE — schema join |
| parallel tools | provider=Anthropic; host=api.anthropic.com; old=claude-sonnet-4-6; new=claude-sonnet-5; setting=tools=parallel; call IDs=required; ordering=observed; evidence=2026-09-01 | tool replay required before changing traffic | CONDITIONAL — tool probe |
| computer-use | provider=Anthropic; host=api.anthropic.com; old=claude-sonnet-4-6; new=claude-sonnet-5; setting=modality=computer-use; actions=contained; response=trace; evidence=2026-09-01 | support=Unknown; do not assume version parity | UNAVAILABLE — modality/control |
| long context | provider=Anthropic; host=api.anthropic.com; old=claude-sonnet-4-6; new=claude-sonnet-5; setting=input=long; output reserve=declared; cache=separate; evidence=2026-09-01 | cache/token consequence=Unavailable without exact accounting | UNAVAILABLE — cache/context join |
Provenance: claude-sonnet-4-6-vs-claude-sonnet-5, first-party evidence, surface verification date 2026-09-01. Anthropic Claude model documentation. Missing joins fail closed.
Version-normalized Sonnet evidence bridge
Frozen fixture board. Formula / decision rule: delta = new − old only for same evaluation/version/settings; otherwise Unavailable Boundary: Provider launch evidence is labeled separately from local canary acceptance evidence.
| Frozen fixture | Joined fields and evidence | Output | State |
|---|---|---|---|
| same benchmark and version | provider=Anthropic; host=api.anthropic.com; old=claude-sonnet-4-6; new=claude-sonnet-5; setting=evaluation=v; settings=matched; tools=matched; dates=dated; evidence=2026-09-01 | delta=Unavailable until both values and denominator join | CONDITIONAL — comparable shape |
| revised benchmark | provider=Anthropic; host=api.anthropic.com; old=claude-sonnet-4-6; new=claude-sonnet-5; setting=evaluation=revA/revB; settings=unknown; evidence=2026-09-01 | do not compute generation delta | NOT COMPARABLE — revision changed |
| changed effort | provider=Anthropic; host=api.anthropic.com; old=claude-sonnet-4-6; new=claude-sonnet-5; setting=evaluation=v; old effort=medium; new effort=high; evidence=2026-09-01 | exclude from model-only upgrade delta | NOT COMPARABLE — effort changed |
| changed tool policy | provider=Anthropic; host=api.anthropic.com; old=claude-sonnet-4-6; new=claude-sonnet-5; setting=evaluation=v; old tools=none; new tools=search; evidence=2026-09-01 | tool delta is a confounder | NOT COMPARABLE — tool policy changed |
| provider-claimed score | provider=Anthropic; host=api.anthropic.com; old=claude-sonnet-4-6; new=claude-sonnet-5; setting=source=provider; score=claim; judge/denominator=Unavailable; evidence=2026-09-01 | launch evidence can trigger a probe, not production acceptance | PROVIDER CLAIM — probe required |
| local canary | provider=Anthropic; host=api.anthropic.com; old=claude-sonnet-4-6; new=claude-sonnet-5; setting=source=local; prompt/config hash=paired; rubric=declared; sample_floor=required; evidence=2026-09-01 | promotion=Unavailable until sample floor and critical gates pass | UNAVAILABLE — canary incomplete |
Provenance: claude-sonnet-4-6-vs-claude-sonnet-5, first-party evidence, surface verification date 2026-09-01. Anthropic Claude model documentation. Missing joins fail closed.
Sonnet upgrade canary ledger
Frozen fixture board. Formula / decision rule: promote = every critical cohort passes quality/tool/schema/latency gates; rollback target remains 4.6 until then Boundary: Canary decisions use workload observations, not launch benchmarks or a renamed generic comparison.
| Frozen fixture | Joined fields and evidence | Output | State |
|---|---|---|---|
| read-only chat | cohort=chat; baseline_hash=4.6; candidate_hash=5; tools=none; quality_rubric=declared; sample_floor=required | decision=Hold until paired sample passes | HOLD — baseline retained |
| structured extraction | cohort=extract; schema_hash=paired; field_agreement=measured required; cache=separate | rollback if schema or field gate regresses | GATE — schema acceptance |
| coding edit | cohort=code; baseline_hash=paired; tests=required; tool=repository; latency/token delta=measured | promote only after tests and SLO pass | CONDITIONAL — coding gate |
| parallel-tool agent | cohort=agent; tool IDs/order=paired; side_effects=contained; attempt_cap=declared | hold on tool mismatch or side-effect uncertainty | HOLD — tool containment |
| computer-use task | cohort=computer; action trace=required; human gate=required; support=Unknown | decision=Unavailable until capability and safety controls join | UNAVAILABLE — control evidence |
| long-context workflow | cohort=long; prompt_hash=paired; cache_state=recorded; output reserve=declared; 4.6 rollback alive | promote only when cache and quality deltas are measured | CONDITIONAL — long-context gate |
Provenance: claude-sonnet-4-6-vs-claude-sonnet-5, first-party evidence, surface verification date 2026-09-01. Anthropic Claude model documentation. Missing joins fail closed.
Method and limitations: calculations use only the displayed deterministic rule and visibly labeled assumptions. Missing provider, host, alias, version, effort, tool, modality, benchmark, workload, rate-period, or date joins fail closed. Run this scenario →
Exact-pair decision guide · verified 2026-09-07
Exact pair boundary: Anthropic; same provider, identical base tariffs ($3/$15), version generation upgrade, and tool performance must join. Requested models are claude-sonnet-4-6 and claude-sonnet-5. Provider, rate, pricing, and task pages remain fact owners.
Zero-price-delta migration economic ledger
Frozen scenario board. Formula / deterministic rule: cost_delta = sonnet5_bill - sonnet46_bill = 0; upgrade incentive is pure capability uplift Boundary: Owns zero-cost migration mechanics and validation gates from Sonnet 4.6 to Sonnet 5.
| Frozen scenario | Pair, identity, host, surface, and evidence fields | Result | State |
|---|---|---|---|
| 10K interactive chat queries ($3/$15) | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=10K interactive chat queries ($3/$15); workload volume; Sonnet 4.6 cost; Sonnet 5 cost; financial variance; quality uplift expectation; migration recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 10K interactive chat queries ($3/$15) is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 50K document extraction requests | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=50K document extraction requests; workload volume; Sonnet 4.6 cost; Sonnet 5 cost; financial variance; quality uplift expectation; migration recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 50K document extraction requests is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 100K enterprise agent tool calls | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=100K enterprise agent tool calls; workload volume; Sonnet 4.6 cost; Sonnet 5 cost; financial variance; quality uplift expectation; migration recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 100K enterprise agent tool calls is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| prompt caching active workload | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=prompt caching active workload; workload volume; Sonnet 4.6 cost; Sonnet 5 cost; financial variance; quality uplift expectation; migration recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — prompt caching active workload is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| batch 24h queue job workload | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=batch 24h queue job workload; workload volume; Sonnet 4.6 cost; Sonnet 5 cost; financial variance; quality uplift expectation; migration recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — batch 24h queue job workload is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| unresolved parameter compatibility | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=unresolved parameter compatibility; workload volume; Sonnet 4.6 cost; Sonnet 5 cost; financial variance; quality uplift expectation; migration recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — unresolved parameter compatibility has no matched, dated bilateral observation. | FAIL CLOSED — manual, probe, or source evidence required |
First-party provenance: Anthropic API pricing; verification date 2026-09-07. Missing or conflicting joins fail closed.
Agent tool call success rate and repair cost savings
Frozen scenario board. Formula / deterministic rule: repair_savings = error_reduction_rate * calls * repair_cost_per_failure Boundary: Owns tool call reliability and retry cost reductions.
| Frozen scenario | Pair, identity, host, surface, and evidence fields | Result | State |
|---|---|---|---|
| simple single-tool execution | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=simple single-tool execution; tool complexity; baseline failure rate; Sonnet 5 failure rate; failures prevented; monthly retry savings; reliability verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — simple single-tool execution is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| nested multi-tool dependencies | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=nested multi-tool dependencies; tool complexity; baseline failure rate; Sonnet 5 failure rate; failures prevented; monthly retry savings; reliability verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — nested multi-tool dependencies is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| parallel tool calling sequence | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=parallel tool calling sequence; tool complexity; baseline failure rate; Sonnet 5 failure rate; failures prevented; monthly retry savings; reliability verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — parallel tool calling sequence is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| long-horizon browser agent workflow | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=long-horizon browser agent workflow; tool complexity; baseline failure rate; Sonnet 5 failure rate; failures prevented; monthly retry savings; reliability verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — long-horizon browser agent workflow is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| schema validation retry boundary | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=schema validation retry boundary; tool complexity; baseline failure rate; Sonnet 5 failure rate; failures prevented; monthly retry savings; reliability verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — schema validation retry boundary is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| unhandled tool exception scenario | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=unhandled tool exception scenario; tool complexity; baseline failure rate; Sonnet 5 failure rate; failures prevented; monthly retry savings; reliability verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — unhandled tool exception scenario is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
First-party provenance: Anthropic Claude model documentation; verification date 2026-09-07. Missing or conflicting joins fail closed.
Enterprise rollout canary and parameter parity audit
Frozen scenario board. Formula / deterministic rule: rollout_stage = traffic_pct <= risk_threshold && latency_delta <= max_allowed_ms Boundary: Owns staged enterprise rollout and parameter compatibility validation.
| Frozen scenario | Pair, identity, host, surface, and evidence fields | Result | State |
|---|---|---|---|
| 1% shadow traffic test | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=1% shadow traffic test; traffic allocation; sample volume; latency difference; format agreement; rollback status; migration gate; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 1% shadow traffic test is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 10% early canary validation | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=10% early canary validation; traffic allocation; sample volume; latency difference; format agreement; rollback status; migration gate; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 10% early canary validation is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 50% phased team rollout | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=50% phased team rollout; traffic allocation; sample volume; latency difference; format agreement; rollback status; migration gate; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 50% phased team rollout is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 100% full production migration | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=100% full production migration; traffic allocation; sample volume; latency difference; format agreement; rollback status; migration gate; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 100% full production migration is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| latency regression rollback trigger | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=latency regression rollback trigger; traffic allocation; sample volume; latency difference; format agreement; rollback status; migration gate; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — latency regression rollback trigger is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| divergent output format conflict | pair=claude-sonnet-4-6 vs claude-sonnet-5; provider=Anthropic; hostA=api.anthropic.com; hostB=api.anthropic.com; surfaceA=Anthropic Messages API; surfaceB=Anthropic Messages API; requested/effective IDs=claude-sonnet-4-6,claude-sonnet-5; scenario=divergent output format conflict; traffic allocation; sample volume; latency difference; format agreement; rollback status; migration gate; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — divergent output format conflict is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
First-party provenance: Anthropic API rate limits; verification date 2026-09-07. Missing or conflicting joins fail closed.
Method and limitations: formulas are deterministic; observed and assumed inputs are labeled; no missing provider, host, account, region, realm, alias, snapshot, revision, weight, artifact, control, tool, modality, workload, rate-period, timestamp, or result is transferred. Run this scenario →
How do Claude Sonnet 4.6 and Claude Sonnet 5 compare on specs?
| Claude Sonnet 4.6 | Claude Sonnet 5 | |
|---|---|---|
| Price (input) | $3.00/M | $2.00/M ✓ |
| Price (output) | $15.00/M | $10.00/M ✓ |
| Blended price | $6.00/M | $4.00/M ✓ |
| Context window | 300,000 tokens | 500,000 tokens ✓ |
| Max output | 64,000 tokens | 64,000 tokens |
| Modalities | text, vision | text, vision |
| Reasoning mode | Yes | Yes |
| Released | 2026-03 | 2026-06 |
| Speed | 76 t/s | Not measured |
How much do Claude Sonnet 4.6 and Claude Sonnet 5 cost at scale?
| Tokens / month | Claude Sonnet 4.6 | Claude Sonnet 5 | Delta |
|---|---|---|---|
| 1,000,000 | $6.00 | $4.00 | $2.00 (1.5×) |
| 10,000,000 | $60.00 | $40.00 | $20.00 (1.5×) |
| 100,000,000 | $600.00 | $400.00 | $200.00 (1.5×) |
Choose Claude Sonnet 4.6 if…
- ✓Best intelligence-per-dollar in the Claude lineup
- ✓Fast enough for interactive coding agents
- ✓Reliable structured output
- ✓Enterprise workloads that need Opus-adjacent quality at Sonnet pricing.
Choose Claude Sonnet 5 if…
- ✓Frontier intelligence at Sonnet economics
- ✓Fast interactive coding performance
- ✓Reliable structured output and extended thinking
- ✓Production workloads that need near-Opus quality with better speed and cost.
Run this exact matchup right now
Send the same prompt to Claude Sonnet 4.6 and Claude Sonnet 5 side by side and see the outputs yourself.
Try Claude Sonnet 4.6 vs Claude Sonnet 5 FreeWhat are common questions about Claude Sonnet 4.6 and Claude Sonnet 5?
Is Claude Sonnet 4.6 cheaper than Claude Sonnet 5?
Claude Sonnet 5 is cheaper, at $4.00 per million blended tokens vs $6.00 for Claude Sonnet 4.6.
Which has the bigger context window, Claude Sonnet 4.6 or Claude Sonnet 5?
Claude Sonnet 5 has the larger context window: 500,000 tokens vs 300,000.
Can Claude Sonnet 4.6 replace Claude Sonnet 5 for coding?
Both are viable for coding. Best intelligence-per-dollar in the Claude lineup (Claude Sonnet 4.6) vs Frontier intelligence at Sonnet economics (Claude Sonnet 5) — pick based on which strength matters more for your workload.
Neither of these? See Claude Sonnet 4.6 alternatives or Claude Sonnet 5 alternatives.
What related comparisons help choose between Claude Sonnet 4.6 and Claude Sonnet 5?
Claude Sonnet 4.6 pricingClaude Sonnet 5 pricingvs Claude Sonnet 4.5vs Claude Sonnet 4vs Claude Opus 4.8vs Claude Haiku 4.5Premium model tests
Pricing verified 2026-04-06. Specs verified 2026-08-14.